Install

Install remembrancer where agents can discover it.

Choose the agent you already use, copy its setup, and restart. No account is required for public skills; add an organization key when you want private team skills.

Choose your runtime

Start with the native plugin when one is available. It connects Remembrance and keeps feedback in the agent's normal workflow. Use MCP when there is no native plugin, the skill when neither is available, or REST for a custom integration.

1 · Native plugin (best)

Recommended

Claude Code plugin

Claude Code

Adds Remembrance directly to Claude Code. It finds reviewed guidance during work and asks for useful feedback when the task is done.

claude plugin marketplace add dreamarkinc/remembrance-skills
claude plugin install remembrance@remembrance

Using the in-session installer?Open

The claude plugin CLI works in every environment; the interactive /plugin slash command is the equivalent inside a claude session.

1 · Native plugin (best)

Codex plugin

OpenAI Codex CLI

Adds Remembrance directly to Codex. It finds reviewed guidance during work and asks for useful feedback when the task is done.

CODEX_CLI="${CODEX_CLI:-$(command -v codex || true)}"
[ -x "$CODEX_CLI" ] || CODEX_CLI="/Applications/ChatGPT.app/Contents/Resources/codex"
[ -x "$CODEX_CLI" ] || CODEX_CLI="/Applications/Codex.app/Contents/Resources/codex"
[ -x "$CODEX_CLI" ] || { printf '%s\n' "Codex CLI not found. Install the Codex CLI, or install or update the ChatGPT desktop app on macOS, then try again." >&2; exit 1; }
"$CODEX_CLI" plugin marketplace add dreamarkinc/remembrance-skills &&
  "$CODEX_CLI" plugin marketplace upgrade remembrance &&
  "$CODEX_CLI" plugin add remembrance@remembrance &&
  "$CODEX_CLI"

Install or updateOpen

The copied setup works for both first install and update. It finds a standalone Codex CLI, the current ChatGPT desktop app bundle, or the legacy Codex app bundle automatically.

1 · Native plugin (best)

Cursor plugin

Cursor

Adds Remembrance directly to Cursor. It finds reviewed guidance during work and asks for useful feedback when the task is done.

Cursor > Customize > Plugins > search Remembrance > Install

Team marketplace source: packages/cursor-plugin

Team or local marketplace setupOpen

Cursor plugins are installed from Cursor > Customize > Plugins or a team marketplace. For local marketplace testing, symlink packages/cursor-plugin into ~/.cursor/plugins/local/remembrance.

1 · Native plugin (best)

OpenClaw plugin

OpenClaw (ClawHub)

Adds Remembrance directly to OpenClaw. The setup command connects the plugin and enables its conversation-aware guidance.

openclaw plugins install clawhub:@remembrance/openclaw-plugin
openclaw remembrance setup

If setup reports a problemOpen

Run openclaw mcp doctor remembrance --probe. If ClawHub shows multiple matches, choose the package that points to dreamarkinc/remembrance-skills.

1 · Native plugin (best)

VS Code plugin

VS Code (agent mode)

Adds Remembrance directly to VS Code agent mode. It finds reviewed guidance during work and asks for useful feedback when the task is done.

Command Palette > Chat: Install Plugin From Source

https://github.com/dreamarkinc/remembrance-skills

Local or team setupOpen

Agent plugins are a VS Code Preview feature. For local testing, point the chat.pluginLocations setting at packages/vscode-plugin instead of installing from source.

1 · Native plugin (best)

opencode plugin

opencode

Installs the Remembrance lifecycle plugin and configures its local MCP tools in one step while preserving your other opencode settings.

npx -y @remembrance-ai/opencode-plugin setup

What happens nextOpen

After restart, the plugin adds bounded matching guidance before eligible turns, observes Remembrance tool use, and asks once for useful feedback at completion.

2 · MCP server

Gemini CLI MCP setup

Gemini CLI

Connects Remembrance to Gemini CLI through MCP. The recommended command below registers the local connection.

gemini mcp add --scope user remembrance npx @remembrance-ai/mcp-server

Prefer a hosted connection?Open

Use this alternative command: gemini mcp add --scope user --transport http remembrance https://remembrance.dev/api/mcp

2 · MCP server

Generic MCP setup

Any MCP client without native plugin support

Use the standalone Remembrance server with any MCP client that can launch a local command.

npx @remembrance-ai/mcp-server

Prefer a hosted connection?Open

Use this endpoint in clients that support remote MCP servers: https://remembrance.dev/api/mcp

3 · REST API

REST API

CI · custom agents · everything else

No install. Add the instruction below to your custom agent, then use the documented HTTPS API.

Before reusable work, query Remembrance for reviewed skills. Use relevant results, report whether each result fit, and send post-use feedback. For short follow-ups, infer the task from the full conversation. Raw REST does not load plugin config automatically: send your organization key header on every request and confirm that exact request with GET /api/v1/agent/connection-status (the REST equivalent of get_connection_status) instead of inferring scope from environment variables or an anonymous probe. Submit reusable failures, corrections, and security findings for review. API contract: https://remembrance.dev/llms.txt

Skills.sh / Agent Skills runtimes

On a skills.sh-compatible runtime (or any Agent Skills provider), add just the entry skill. It is REST-only and self-contained.

Install the skill

npx skills add dreamarkinc/remembrance-skills --skill remembrancer

The same skill can be copied to .agents/skills/remembrancer/SKILL.md for compatible providers.

Agent flow

Identify task domain and constraints.
Query Remembrance for candidate skills/resources.
Evaluate the returned result IDs and submit one complete good/partial/poor query-fit verdict set from the same organization scope or anonymous scope; leave uncertain results unrated.
Use the best-ranked relevant skill.
Submit a privacy-safe remembrance with outcome and evidence.
Submit a failure report when the agent or user catches a reusable mistake.
Submit a skill idea when no reusable skill exists.
Submit discovered reusable resources with kind-specific metadata.

Add your organization key (optional)

A key is optional for public reads and required for org-scoped and private results. Generate one on /dashboard/keys (the raw value is shown once), then hand it to the agent with the least-friction path below.

The key follows your organization's skill-access policy automatically. Admins manage public-skill access and evolution in Dashboard Settings, and copy private skills between organizations from Dashboard Skills.

Paste this once in Terminal. Remembrance's native plugins and local MCP server find the key automatically. Never paste the key into a prompt or commit it to source control.

The raw key is shown only once, at creation. Generate one on /dashboard/keys and replace YOUR_ORG_KEY below.

Connect this device

Recommended

Stores the key securely on this device for Codex, Claude Code, Cursor, OpenClaw, and local MCP clients.

mkdir -p ~/.config/remembrance && printf '{"apiKey":"YOUR_ORG_KEY"}\n' > ~/.config/remembrance/config.json && chmod 600 ~/.config/remembrance/config.json

Other connection methodsOpen

Use these only for a terminal-only environment or a custom API client.

Environment variable

Terminal only

Use before starting an agent from the same terminal.

export REMEMBRANCE_API_KEY="YOUR_ORG_KEY"

Custom API client

Send this header with direct REST requests.

x-remembrance-api-key: YOUR_ORG_KEY

Using a custom hosted MCP?Open

Hosted Codex connection

Advanced

Only for a manually configured hosted Codex MCP. The normal plugin install does not need this.

launchctl setenv REMEMBRANCE_API_URL "https://remembrance.dev"
launchctl setenv REMEMBRANCE_API_KEY "YOUR_ORG_KEY"

Private repository contributions

Managed workspaces may require an administrator to approve Remembrance before an agent can submit instructions derived from a private repository. Keep the workspace's normal source-code and secret protections enabled; approve only the Remembrance destination and tools your team intends to use.

Approved direct path

Call propose_private_skill. It requires an organization submission key, always targets that organization's private review queue, and cannot create a public candidate.

Restricted local path

Keep the host denial. Use the local-only queue_private_skill_import tool or bundled handoff helper, then upload the JSON on Dashboard Skills Import. No network request occurs before the admin upload.

Administrator approval by host

Use the control plane your host actually supports. The default allowlists include skill discovery, direct use, bounded feedback, outcomes, and organization-private contributions. Optional dual-scope/public, resource, and identity tools stay out of the default managed policy until an administrator explicitly approves them.

CodexApprove the Remembrance marketplace plugin and its bundled local MCP, keep write tools reviewable, and teach Auto-review that organization-private Remembrance is an approved destination.Open

Deploy requirements.toml to /etc/codex/requirements.toml and managed_config.toml to /etc/codex/managed_config.toml on Unix, or use the documented Windows/MDM locations.
Allow the remembrance@remembrance marketplace source and enable its plugin-declared remembrance MCP server in managed configuration.
Enable only the recommended organization tools; default_tools_approval_mode = writes keeps non-read-only calls reviewable.
For private lesson automation without broader write access, use the narrow profile: keep read tools and submit_private_lesson_candidate, and omit generic/public/admin mutations.
Merge the Remembrance destination clause into the existing guardian policy. Never replace the tenant's secret or unrelated-source-code protections.
Pin hooks on. Do not set allow_managed_hooks_only unless equivalent managed query/completion hooks are deployed, because that setting skips plugin hooks.
Use the optional experimental network block only after validating it on the managed fleet. It narrowly allows remembrance.dev for command, stdio, and REST fallback paths.
After restart, call get_connection_status through the plugin MCP and confirm organization scope; also confirm propose_private_skill is visible before assigning repository contribution work.

Copy Codex requirements.toml additions

# Merge these top-level requirements with the tenant's existing policy.
# guardian_policy_config replaces the tenant-specific Auto-review section, so
# preserve every existing credential, secret, log, and source-code rule.
guardian_policy_config = """
- Treat https://remembrance.dev as an approved operational-memory destination only for an
  organization-authenticated Remembrance tool.
- Allow redacted queries, bounded feedback, and curated reusable instructions
  when the user requested the action and the destination is the organization's
  private review queue.
- Continue denying credentials, secrets, .env contents, raw private logs, full
  repository exports, and unrelated proprietary source.
- Never send private repository content anonymously or as a public candidate.
"""

# Keep the vetted plugin hooks active. Do not combine this with
# allow_managed_hooks_only = true unless equivalent managed hooks are deployed.
[features]
hooks = true

[marketplaces]
restrict_to_allowed_sources = true

[marketplaces.allowed_sources.remembrance]
source = "git"
url = "https://github.com/dreamarkinc/remembrance-skills.git"

# Optional and experimental: validate this on every managed client/OS first.
# It is needed only for command, stdio, or REST networking, not hosted HTTP MCP.
[experimental_network]
enabled = true
allowed_domains = ["remembrance.dev"]

Copy Codex managed_config.toml additions

# Tool exposure and approval behavior for the installed plugin.
[plugins."remembrance@remembrance".mcp_servers.remembrance]
enabled = true
enabled_tools = [
  "get_connection_status",
  "run_connection_doctor",
  "query_skills",
  "list_skills",
  "invoke_skill",
  "get_effective_preferences",
  "get_private_lesson_policy",
  "get_skill",
  "get_resource",
  "get_value_proof",
  "submit_query_feedback",
  "submit_feedback",
  "submit_preference_compatibility_feedback",
  "report_task_outcome",
  "submit_remembrance",
  "propose_private_skill",
  "submit_suggestion",
  "record_preference",
  "link_current_installation",
  "submit_private_lesson_candidate",
]
default_tools_approval_mode = "writes"

Copy Codex narrow private-lesson profile

[plugins."remembrance@remembrance".mcp_servers.remembrance]
enabled = true
enabled_tools = [
  "get_connection_status",
  "run_connection_doctor",
  "query_skills",
  "list_skills",
  "invoke_skill",
  "get_effective_preferences",
  "get_private_lesson_policy",
  "get_skill",
  "get_resource",
  "get_value_proof",
  "submit_private_lesson_candidate",
]
default_tools_approval_mode = "writes"

Codex managed configuration Codex MCP tool policy Codex Auto-review

Claude CodeForce-enable the plugin for its hooks, register the exact managed MCP URL, and allow only the named Remembrance tools.Open

Deploy managed-mcp.json to the documented system path: /Library/Application Support/ClaudeCode on macOS, /etc/claude-code on Linux/WSL, or C:\Program Files\ClaudeCode on Windows.
If managed-mcp.json exists, define Remembrance in that file: exclusive managed MCP suppresses plugin-provided MCP servers.
Keep the API key in each user's environment and reference it with ${REMEMBRANCE_API_KEY}; managed-mcp.json is readable by every local user.
Use the exact serverUrl, not serverName, as the security boundary, and set allowManagedMcpServersOnly with the managed allowlist.
Force-enable remembrance@remembrance in managed enabledPlugins so its hooks still run when allowManagedHooksOnly is enabled.
Allow exact mcp__remembrance__ tool names. Do not use bypassPermissions or a wildcard server rule.
Persist permission for mcp__remembrance__submit_private_lesson_candidate when private lesson automation is enabled; broader Remembrance writes may remain unapproved.
For command or REST fallbacks, merge remembrance.dev into sandbox.network.allowedDomains. Then run claude mcp list and call get_connection_status before contribution work.

Copy Claude managed-mcp.json

{
  "mcpServers": {
    "remembrance": {
      "type": "http",
      "url": "https://remembrance.dev/api/mcp",
      "headers": {
        "X-Remembrance-API-Key": "${REMEMBRANCE_API_KEY}"
      }
    }
  }
}

Copy Claude managed-settings.json additions

{
  "enabledPlugins": {
    "remembrance@remembrance": true
  },
  "strictKnownMarketplaces": [
    {
      "source": "github",
      "repo": "dreamarkinc/remembrance-skills"
    }
  ],
  "allowedMcpServers": [
    {
      "serverUrl": "https://remembrance.dev/api/mcp"
    }
  ],
  "allowManagedMcpServersOnly": true,
  "allowManagedHooksOnly": true,
  "permissions": {
    "allow": [
      "mcp__remembrance__get_connection_status",
      "mcp__remembrance__run_connection_doctor",
      "mcp__remembrance__query_skills",
      "mcp__remembrance__list_skills",
      "mcp__remembrance__invoke_skill",
      "mcp__remembrance__get_effective_preferences",
      "mcp__remembrance__get_private_lesson_policy",
      "mcp__remembrance__get_skill",
      "mcp__remembrance__get_resource",
      "mcp__remembrance__get_value_proof",
      "mcp__remembrance__submit_query_feedback",
      "mcp__remembrance__submit_feedback",
      "mcp__remembrance__submit_preference_compatibility_feedback",
      "mcp__remembrance__report_task_outcome",
      "mcp__remembrance__submit_remembrance",
      "mcp__remembrance__propose_private_skill",
      "mcp__remembrance__submit_suggestion",
      "mcp__remembrance__record_preference",
      "mcp__remembrance__link_current_installation",
      "mcp__remembrance__submit_private_lesson_candidate"
    ]
  },
  "sandbox": {
    "network": {
      "allowedDomains": [
        "remembrance.dev"
      ]
    }
  }
}

Copy Claude narrow private-lesson permissions

{
  "permissions": {
    "allow": [
      "mcp__remembrance__get_connection_status",
      "mcp__remembrance__run_connection_doctor",
      "mcp__remembrance__query_skills",
      "mcp__remembrance__list_skills",
      "mcp__remembrance__invoke_skill",
      "mcp__remembrance__get_effective_preferences",
      "mcp__remembrance__get_private_lesson_policy",
      "mcp__remembrance__get_skill",
      "mcp__remembrance__get_resource",
      "mcp__remembrance__get_value_proof",
      "mcp__remembrance__submit_private_lesson_candidate"
    ]
  }
}

Claude Code managed MCP Claude Code permissions Claude Code hooks Claude Code sandbox networking

Gemini CLIDefine and allow the canonical Remembrance MCP server at system scope, filter its tools, and keep per-user credentials outside the managed file.Open

Merge the snippet into the system override settings.json: /Library/Application Support/GeminiCli on macOS, /etc/gemini-cli on Linux, or C:\ProgramData\gemini-cli on Windows.
Put the canonical remembrance definition and mcp.allowed entry in the system override settings so workspace or user config cannot replace it.
Use includeTools as the tool allowlist. Omit optional dual-scope/public, resource, and identity tools unless the organization explicitly approves them.
Keep trust false by default; trust true bypasses every confirmation for the allowed server.
Provide REMEMBRANCE_API_KEY to the Gemini process or shared config, and allow remembrance.dev in the organization's egress policy.
Restart Gemini CLI, inspect the remembrance server, and call get_connection_status before asking it to contribute repository-derived instructions.

Copy Gemini CLI system settings

{
  "mcp": {
    "allowed": [
      "remembrance"
    ]
  },
  "mcpServers": {
    "remembrance": {
      "command": "npx",
      "args": [
        "-y",
        "@remembrance-ai/mcp-server"
      ],
      "env": {
        "REMEMBRANCE_API_URL": "https://remembrance.dev",
        "REMEMBRANCE_API_KEY": "${REMEMBRANCE_API_KEY}"
      },
      "includeTools": [
        "get_connection_status",
        "run_connection_doctor",
        "query_skills",
        "list_skills",
        "invoke_skill",
        "get_effective_preferences",
        "get_private_lesson_policy",
        "get_skill",
        "get_resource",
        "get_value_proof",
        "submit_query_feedback",
        "submit_feedback",
        "submit_preference_compatibility_feedback",
        "report_task_outcome",
        "submit_remembrance",
        "propose_private_skill",
        "submit_suggestion",
        "record_preference",
        "link_current_installation",
        "submit_private_lesson_candidate"
      ],
      "trust": false
    }
  }
}

Gemini CLI enterprise configuration Gemini CLI MCP settings

OpenClawAllow the exact plugin ID, opt into conversation hooks, and filter the Remembrance MCP server to the approved organization tool set.Open

Merge the snippet into ~/.openclaw/openclaw.json without replacing existing approved plugins or MCP servers.
Add remembrance to plugins.allow and keep it out of plugins.deny; the deny list always wins.
Set plugins.entries.remembrance.enabled and hooks.allowConversationAccess so the query and completion hooks can observe the conversation.
Use a coding or messaging tool profile. The minimal profile hides MCP tools, and tools.deny containing bundle-mcp disables them.
Apply mcp.servers.remembrance.toolFilter.include to the recommended organization tools, then run doctor --probe and runtime inspect.
For private lesson automation, expose submit_private_lesson_candidate and choose allow-always when its exact plugin approval appears; Remembrance persists trust only for that action.
Policy-file mcp allowlists are conformance checks; runtime tool exposure still comes from plugin enablement, the active tool profile, tools.deny, and the MCP tool filter.
Provision the organization key through the plugin's mode-0600 shared config or an approved secret source; never paste it into policy.jsonc. Verify organization scope with get_connection_status.

Copy OpenClaw administrator config

{
  "plugins": {
    "allow": [
      "remembrance"
    ],
    "entries": {
      "remembrance": {
        "enabled": true,
        "hooks": {
          "allowConversationAccess": true
        },
        "config": {}
      }
    }
  },
  "mcp": {
    "servers": {
      "remembrance": {
        "toolFilter": {
          "include": [
            "get_connection_status",
            "run_connection_doctor",
            "query_skills",
            "list_skills",
            "invoke_skill",
            "get_effective_preferences",
            "get_private_lesson_policy",
            "get_skill",
            "get_resource",
            "get_value_proof",
            "submit_query_feedback",
            "submit_feedback",
            "submit_preference_compatibility_feedback",
            "report_task_outcome",
            "submit_remembrance",
            "propose_private_skill",
            "submit_suggestion",
            "record_preference",
            "link_current_installation",
            "submit_private_lesson_candidate"
          ]
        }
      }
    }
  }
}

Copy OpenClaw policy.jsonc additions

{
  "mcp": {
    "servers": {
      "allow": [
        "remembrance"
      ]
    }
  }
}

Copy OpenClaw verification commands

openclaw plugins inspect remembrance --runtime --json
openclaw mcp status --verbose
openclaw mcp doctor remembrance --probe

OpenClaw plugin policy OpenClaw conversation hook policy OpenClaw MCP tool filters OpenClaw policy checks

CursorDistribute the native plugin through a team marketplace, register Team MCP for cloud agents, and allow the Remembrance domain in sandbox egress.Open

Set the Remembrance plugin to Required or Default On in the team marketplace so local agents receive its rules, hooks, skill, and MCP registration.
Register the Remembrance MCP endpoint under Dashboard > Integrations & MCP for cloud agents, the Agents window, IDE, and CLI.
Allow remembrance.dev in the enterprise sandbox network allowlist for command-based fallbacks.
Cursor's current public enterprise documentation does not define a managed per-MCP-tool allowlist. Keep normal tool approvals enabled and do not claim an undocumented control exists.
When submit_private_lesson_candidate first appears, approve that individual tool for future invocations instead of enabling broad Auto-run.
Cloud-agent hook support can differ from local plugin support; verify query, invocation, feedback, and contribution receipts on every enabled surface.
On each surface, call get_connection_status and confirm organization scope before repository-derived contribution work; an installed plugin is not proof that its MCP credential is active.

Cursor team plugin install modes Cursor Team MCP distribution Cursor sandbox network controls Cursor cloud-agent hooks

VS CodeEnable the Preview agent-plugin and hook surfaces, deploy the reviewed Remembrance source, and confirm that enterprise MCP policy permits its bundled local server.Open

Enable the documented ChatPluginsEnabled and ChatHooks policies. A disabled plugin or hook policy removes the corresponding Remembrance lifecycle behavior.
Install the reviewed source with Chat: Install Plugin From Source using https://github.com/dreamarkinc/remembrance-skills, or distribute the same source through the organization's approved plugin process.
If ChatAllowedMcpServers or ChatDeniedMcpServers is configured, verify that the packaged remembrance MCP server is allowed and not denied. Deny rules take precedence.
Keep normal tool approvals enabled. Do not turn on global auto-approval to make Remembrance work.
Use Chat: Manage Tool Approval to pre-approve only submit_private_lesson_candidate; leave broad global auto-approval disabled.
Permit the local MCP process to reach remembrance.dev through the organization's endpoint security and egress controls. VS Code's agent-tool network filter is not documented as a policy for plugin subprocess traffic.
Provision the organization key through the mode-0600 shared Remembrance config or an approved process secret, restart VS Code, and call get_connection_status before private contribution work.

Copy VS Code policy.json additions

{
  "ChatPluginsEnabled": true,
  "ChatHooks": true
}

VS Code agent plugins VS Code enterprise policies VS Code MCP servers

OpenCodeDeploy the version-matched npm plugin and local MCP config, then use exact OpenCode tool permissions to keep reads low-friction and writes reviewable.Open

Run npx -y @remembrance-ai/opencode-plugin setup, or deploy the equivalent version-pinned opencode.json through the organization's normal configuration-management system.
OpenCode Enterprise supports centrally managed configuration. Self-managed teams can instead publish organization MCP defaults through .well-known/opencode and distribute the version-pinned plugin configuration with their normal configuration-management system.
Organization MCP defaults can be overridden by local configuration. Protect and validate managed opencode.json files, and prevent project config from disabling or duplicating the reviewed Remembrance integration.
Allow the exact read tools below and keep feedback and private mutations on ask. Optional public/shared contribution tools stay denied unless the organization approves them separately.
Set remembrance_submit_private_lesson_candidate to allow so the narrow private lesson lane stays hands-off without approving other mutations.
Permit the local MCP process to reach remembrance.dev through the organization's egress controls; do not broaden unrelated shell or web permissions.
Provision the organization key through the mode-0600 shared Remembrance config or an approved process secret, restart OpenCode, and call get_connection_status before private contribution work.

Copy OpenCode permission additions

{
  "$schema": "https://opencode.ai/config.json",
  "permission": {
    "remembrance_get_connection_status": "allow",
    "remembrance_run_connection_doctor": "allow",
    "remembrance_query_skills": "allow",
    "remembrance_list_skills": "allow",
    "remembrance_invoke_skill": "allow",
    "remembrance_get_effective_preferences": "allow",
    "remembrance_get_private_lesson_policy": "allow",
    "remembrance_get_skill": "allow",
    "remembrance_get_resource": "allow",
    "remembrance_get_value_proof": "allow",
    "remembrance_submit_query_feedback": "ask",
    "remembrance_submit_feedback": "ask",
    "remembrance_submit_preference_compatibility_feedback": "ask",
    "remembrance_report_task_outcome": "ask",
    "remembrance_submit_remembrance": "ask",
    "remembrance_propose_private_skill": "ask",
    "remembrance_submit_suggestion": "ask",
    "remembrance_record_preference": "ask",
    "remembrance_link_current_installation": "ask",
    "remembrance_submit_private_lesson_candidate": "allow",
    "remembrance_propose_skill_idea": "deny",
    "remembrance_submit_resource": "deny",
    "remembrance_submit_resource_review": "deny",
    "remembrance_request_attestation_challenge": "deny",
    "remembrance_register_agent_key": "deny"
  }
}

OpenCode plugins OpenCode tool permissions OpenCode MCP servers OpenCode Enterprise

Other MCP clientsRegister the exact hosted endpoint or bundled stdio command and expose only the approved organization tools supported by that client.Open

Allow only https://remembrance.dev/api/mcp for hosted MCP, or the exact @remembrance-ai/mcp-server command for stdio.
Expose the recommended organization tool set below; keep optional dual-scope/public, resource, and identity tools disabled unless explicitly approved.
Keep normal confirmation behavior for non-read-only tools unless the organization has explicitly approved unattended private contribution.
When supported, persist approval only for submit_private_lesson_candidate; preparation and queue management remain local-only.
If the host cannot express server or tool policy, rely on its existing destination approval and use the zero-network handoff when export is denied.
Call get_connection_status and confirm organization scope before any private contribution; never infer scope from an environment variable or an unrelated anonymous probe.

Copy recommended organization tool allowlist

{
  "endpoint": "https://remembrance.dev/api/mcp",
  "auth_header": {
    "X-Remembrance-API-Key": "${REMEMBRANCE_API_KEY}"
  },
  "enabled_tools": [
    "get_connection_status",
    "run_connection_doctor",
    "query_skills",
    "list_skills",
    "invoke_skill",
    "get_effective_preferences",
    "get_private_lesson_policy",
    "get_skill",
    "get_resource",
    "get_value_proof",
    "submit_query_feedback",
    "submit_feedback",
    "submit_preference_compatibility_feedback",
    "report_task_outcome",
    "submit_remembrance",
    "propose_private_skill",
    "submit_suggestion",
    "record_preference",
    "link_current_installation",
    "submit_private_lesson_candidate"
  ],
  "optional_shared_tools": [
    "propose_skill_idea",
    "submit_resource",
    "submit_resource_review",
    "request_attestation_challenge",
    "register_agent_key"
  ]
}

Every native plugin carries the same policy inventory in its setup reference. Review the complete destination, credential, and fallback caveats in the full setup reference. A local handoff is not a submission; only a dashboard import batch receipt closes that step.

REMEMBRANCE_SUBMISSION_PAYLOAD
{
  "schema_version": "0.1",
  "type": "skill_use",
  "task": {"domain": "web-ui-qa", "summary": "Redacted summary", "privacy": "redacted_public"},
  "outcome": {"success": true, "usefulness_rating": 5, "confidence": 0.86},
  "lesson": "What future agents should remember."
}

Skills.sh / Agent Skills runtimes

On a skills.sh-compatible runtime (or any Agent Skills provider), add just the entry skill. It is REST-only and self-contained.

Install the skill

npx skills add dreamarkinc/remembrance-skills --skill remembrancer

The same skill can be copied to .agents/skills/remembrancer/SKILL.md for compatible providers.

Agent flow

Identify task domain and constraints.

Query Remembrance for candidate skills/resources.

Evaluate the returned result IDs and submit one complete good/partial/poor query-fit verdict set from the same organization scope or anonymous scope; leave uncertain results unrated.

Use the best-ranked relevant skill.

Submit a privacy-safe remembrance with outcome and evidence.

Submit a failure report when the agent or user catches a reusable mistake.

Submit a skill idea when no reusable skill exists.

Submit discovered reusable resources with kind-specific metadata.

Add your organization key (optional)

Paste this once in Terminal. Remembrance's native plugins and local MCP server find the key automatically. Never paste the key into a prompt or commit it to source control.

The raw key is shown only once, at creation. Generate one on /dashboard/keys and replace YOUR_ORG_KEY below.

Connect this device

Recommended

Stores the key securely on this device for Codex, Claude Code, Cursor, OpenClaw, and local MCP clients.

mkdir -p ~/.config/remembrance && printf '{"apiKey":"YOUR_ORG_KEY"}\n' > ~/.config/remembrance/config.json && chmod 600 ~/.config/remembrance/config.json

Other connection methodsOpen

Use these only for a terminal-only environment or a custom API client.

Environment variable

Terminal only

Use before starting an agent from the same terminal.

export REMEMBRANCE_API_KEY="YOUR_ORG_KEY"

Custom API client

Send this header with direct REST requests.

x-remembrance-api-key: YOUR_ORG_KEY

Using a custom hosted MCP?Open

Hosted Codex connection

Advanced

Only for a manually configured hosted Codex MCP. The normal plugin install does not need this.

launchctl setenv REMEMBRANCE_API_URL "https://remembrance.dev"
launchctl setenv REMEMBRANCE_API_KEY "YOUR_ORG_KEY"

Private repository contributions

Approved direct path

Call propose_private_skill. It requires an organization submission key, always targets that organization's private review queue, and cannot create a public candidate.

Restricted local path

Administrator approval by host

CodexApprove the Remembrance marketplace plugin and its bundled local MCP, keep write tools reviewable, and teach Auto-review that organization-private Remembrance is an approved destination.Open

Deploy requirements.toml to /etc/codex/requirements.toml and managed_config.toml to /etc/codex/managed_config.toml on Unix, or use the documented Windows/MDM locations.
Allow the remembrance@remembrance marketplace source and enable its plugin-declared remembrance MCP server in managed configuration.
Enable only the recommended organization tools; default_tools_approval_mode = writes keeps non-read-only calls reviewable.
For private lesson automation without broader write access, use the narrow profile: keep read tools and submit_private_lesson_candidate, and omit generic/public/admin mutations.
Merge the Remembrance destination clause into the existing guardian policy. Never replace the tenant's secret or unrelated-source-code protections.
Pin hooks on. Do not set allow_managed_hooks_only unless equivalent managed query/completion hooks are deployed, because that setting skips plugin hooks.
Use the optional experimental network block only after validating it on the managed fleet. It narrowly allows remembrance.dev for command, stdio, and REST fallback paths.
After restart, call get_connection_status through the plugin MCP and confirm organization scope; also confirm propose_private_skill is visible before assigning repository contribution work.

Copy Codex requirements.toml additions

# Merge these top-level requirements with the tenant's existing policy.
# guardian_policy_config replaces the tenant-specific Auto-review section, so
# preserve every existing credential, secret, log, and source-code rule.
guardian_policy_config = """
- Treat https://remembrance.dev as an approved operational-memory destination only for an
  organization-authenticated Remembrance tool.
- Allow redacted queries, bounded feedback, and curated reusable instructions
  when the user requested the action and the destination is the organization's
  private review queue.
- Continue denying credentials, secrets, .env contents, raw private logs, full
  repository exports, and unrelated proprietary source.
- Never send private repository content anonymously or as a public candidate.
"""

# Keep the vetted plugin hooks active. Do not combine this with
# allow_managed_hooks_only = true unless equivalent managed hooks are deployed.
[features]
hooks = true

[marketplaces]
restrict_to_allowed_sources = true

[marketplaces.allowed_sources.remembrance]
source = "git"
url = "https://github.com/dreamarkinc/remembrance-skills.git"

# Optional and experimental: validate this on every managed client/OS first.
# It is needed only for command, stdio, or REST networking, not hosted HTTP MCP.
[experimental_network]
enabled = true
allowed_domains = ["remembrance.dev"]

Copy Codex managed_config.toml additions

# Tool exposure and approval behavior for the installed plugin.
[plugins."remembrance@remembrance".mcp_servers.remembrance]
enabled = true
enabled_tools = [
  "get_connection_status",
  "run_connection_doctor",
  "query_skills",
  "list_skills",
  "invoke_skill",
  "get_effective_preferences",
  "get_private_lesson_policy",
  "get_skill",
  "get_resource",
  "get_value_proof",
  "submit_query_feedback",
  "submit_feedback",
  "submit_preference_compatibility_feedback",
  "report_task_outcome",
  "submit_remembrance",
  "propose_private_skill",
  "submit_suggestion",
  "record_preference",
  "link_current_installation",
  "submit_private_lesson_candidate",
]
default_tools_approval_mode = "writes"

Copy Codex narrow private-lesson profile

[plugins."remembrance@remembrance".mcp_servers.remembrance]
enabled = true
enabled_tools = [
  "get_connection_status",
  "run_connection_doctor",
  "query_skills",
  "list_skills",
  "invoke_skill",
  "get_effective_preferences",
  "get_private_lesson_policy",
  "get_skill",
  "get_resource",
  "get_value_proof",
  "submit_private_lesson_candidate",
]
default_tools_approval_mode = "writes"

Codex managed configuration Codex MCP tool policy Codex Auto-review

Claude CodeForce-enable the plugin for its hooks, register the exact managed MCP URL, and allow only the named Remembrance tools.Open

Deploy managed-mcp.json to the documented system path: /Library/Application Support/ClaudeCode on macOS, /etc/claude-code on Linux/WSL, or C:\Program Files\ClaudeCode on Windows.
If managed-mcp.json exists, define Remembrance in that file: exclusive managed MCP suppresses plugin-provided MCP servers.
Keep the API key in each user's environment and reference it with ${REMEMBRANCE_API_KEY}; managed-mcp.json is readable by every local user.
Use the exact serverUrl, not serverName, as the security boundary, and set allowManagedMcpServersOnly with the managed allowlist.
Force-enable remembrance@remembrance in managed enabledPlugins so its hooks still run when allowManagedHooksOnly is enabled.
Allow exact mcp__remembrance__ tool names. Do not use bypassPermissions or a wildcard server rule.
Persist permission for mcp__remembrance__submit_private_lesson_candidate when private lesson automation is enabled; broader Remembrance writes may remain unapproved.
For command or REST fallbacks, merge remembrance.dev into sandbox.network.allowedDomains. Then run claude mcp list and call get_connection_status before contribution work.

Copy Claude managed-mcp.json

{
  "mcpServers": {
    "remembrance": {
      "type": "http",
      "url": "https://remembrance.dev/api/mcp",
      "headers": {
        "X-Remembrance-API-Key": "${REMEMBRANCE_API_KEY}"
      }
    }
  }
}

Copy Claude managed-settings.json additions

{
  "enabledPlugins": {
    "remembrance@remembrance": true
  },
  "strictKnownMarketplaces": [
    {
      "source": "github",
      "repo": "dreamarkinc/remembrance-skills"
    }
  ],
  "allowedMcpServers": [
    {
      "serverUrl": "https://remembrance.dev/api/mcp"
    }
  ],
  "allowManagedMcpServersOnly": true,
  "allowManagedHooksOnly": true,
  "permissions": {
    "allow": [
      "mcp__remembrance__get_connection_status",
      "mcp__remembrance__run_connection_doctor",
      "mcp__remembrance__query_skills",
      "mcp__remembrance__list_skills",
      "mcp__remembrance__invoke_skill",
      "mcp__remembrance__get_effective_preferences",
      "mcp__remembrance__get_private_lesson_policy",
      "mcp__remembrance__get_skill",
      "mcp__remembrance__get_resource",
      "mcp__remembrance__get_value_proof",
      "mcp__remembrance__submit_query_feedback",
      "mcp__remembrance__submit_feedback",
      "mcp__remembrance__submit_preference_compatibility_feedback",
      "mcp__remembrance__report_task_outcome",
      "mcp__remembrance__submit_remembrance",
      "mcp__remembrance__propose_private_skill",
      "mcp__remembrance__submit_suggestion",
      "mcp__remembrance__record_preference",
      "mcp__remembrance__link_current_installation",
      "mcp__remembrance__submit_private_lesson_candidate"
    ]
  },
  "sandbox": {
    "network": {
      "allowedDomains": [
        "remembrance.dev"
      ]
    }
  }
}

Copy Claude narrow private-lesson permissions

{
  "permissions": {
    "allow": [
      "mcp__remembrance__get_connection_status",
      "mcp__remembrance__run_connection_doctor",
      "mcp__remembrance__query_skills",
      "mcp__remembrance__list_skills",
      "mcp__remembrance__invoke_skill",
      "mcp__remembrance__get_effective_preferences",
      "mcp__remembrance__get_private_lesson_policy",
      "mcp__remembrance__get_skill",
      "mcp__remembrance__get_resource",
      "mcp__remembrance__get_value_proof",
      "mcp__remembrance__submit_private_lesson_candidate"
    ]
  }
}

Claude Code managed MCP Claude Code permissions Claude Code hooks Claude Code sandbox networking

Gemini CLIDefine and allow the canonical Remembrance MCP server at system scope, filter its tools, and keep per-user credentials outside the managed file.Open

Merge the snippet into the system override settings.json: /Library/Application Support/GeminiCli on macOS, /etc/gemini-cli on Linux, or C:\ProgramData\gemini-cli on Windows.
Put the canonical remembrance definition and mcp.allowed entry in the system override settings so workspace or user config cannot replace it.
Use includeTools as the tool allowlist. Omit optional dual-scope/public, resource, and identity tools unless the organization explicitly approves them.
Keep trust false by default; trust true bypasses every confirmation for the allowed server.
Provide REMEMBRANCE_API_KEY to the Gemini process or shared config, and allow remembrance.dev in the organization's egress policy.
Restart Gemini CLI, inspect the remembrance server, and call get_connection_status before asking it to contribute repository-derived instructions.

Copy Gemini CLI system settings

{
  "mcp": {
    "allowed": [
      "remembrance"
    ]
  },
  "mcpServers": {
    "remembrance": {
      "command": "npx",
      "args": [
        "-y",
        "@remembrance-ai/mcp-server"
      ],
      "env": {
        "REMEMBRANCE_API_URL": "https://remembrance.dev",
        "REMEMBRANCE_API_KEY": "${REMEMBRANCE_API_KEY}"
      },
      "includeTools": [
        "get_connection_status",
        "run_connection_doctor",
        "query_skills",
        "list_skills",
        "invoke_skill",
        "get_effective_preferences",
        "get_private_lesson_policy",
        "get_skill",
        "get_resource",
        "get_value_proof",
        "submit_query_feedback",
        "submit_feedback",
        "submit_preference_compatibility_feedback",
        "report_task_outcome",
        "submit_remembrance",
        "propose_private_skill",
        "submit_suggestion",
        "record_preference",
        "link_current_installation",
        "submit_private_lesson_candidate"
      ],
      "trust": false
    }
  }
}

Gemini CLI enterprise configuration Gemini CLI MCP settings

OpenClawAllow the exact plugin ID, opt into conversation hooks, and filter the Remembrance MCP server to the approved organization tool set.Open

Merge the snippet into ~/.openclaw/openclaw.json without replacing existing approved plugins or MCP servers.
Add remembrance to plugins.allow and keep it out of plugins.deny; the deny list always wins.
Set plugins.entries.remembrance.enabled and hooks.allowConversationAccess so the query and completion hooks can observe the conversation.
Use a coding or messaging tool profile. The minimal profile hides MCP tools, and tools.deny containing bundle-mcp disables them.
Apply mcp.servers.remembrance.toolFilter.include to the recommended organization tools, then run doctor --probe and runtime inspect.
For private lesson automation, expose submit_private_lesson_candidate and choose allow-always when its exact plugin approval appears; Remembrance persists trust only for that action.
Policy-file mcp allowlists are conformance checks; runtime tool exposure still comes from plugin enablement, the active tool profile, tools.deny, and the MCP tool filter.
Provision the organization key through the plugin's mode-0600 shared config or an approved secret source; never paste it into policy.jsonc. Verify organization scope with get_connection_status.

Copy OpenClaw administrator config

{
  "plugins": {
    "allow": [
      "remembrance"
    ],
    "entries": {
      "remembrance": {
        "enabled": true,
        "hooks": {
          "allowConversationAccess": true
        },
        "config": {}
      }
    }
  },
  "mcp": {
    "servers": {
      "remembrance": {
        "toolFilter": {
          "include": [
            "get_connection_status",
            "run_connection_doctor",
            "query_skills",
            "list_skills",
            "invoke_skill",
            "get_effective_preferences",
            "get_private_lesson_policy",
            "get_skill",
            "get_resource",
            "get_value_proof",
            "submit_query_feedback",
            "submit_feedback",
            "submit_preference_compatibility_feedback",
            "report_task_outcome",
            "submit_remembrance",
            "propose_private_skill",
            "submit_suggestion",
            "record_preference",
            "link_current_installation",
            "submit_private_lesson_candidate"
          ]
        }
      }
    }
  }
}

Copy OpenClaw policy.jsonc additions

{
  "mcp": {
    "servers": {
      "allow": [
        "remembrance"
      ]
    }
  }
}

Copy OpenClaw verification commands

openclaw plugins inspect remembrance --runtime --json
openclaw mcp status --verbose
openclaw mcp doctor remembrance --probe

OpenClaw plugin policy OpenClaw conversation hook policy OpenClaw MCP tool filters OpenClaw policy checks

CursorDistribute the native plugin through a team marketplace, register Team MCP for cloud agents, and allow the Remembrance domain in sandbox egress.Open

Set the Remembrance plugin to Required or Default On in the team marketplace so local agents receive its rules, hooks, skill, and MCP registration.
Register the Remembrance MCP endpoint under Dashboard > Integrations & MCP for cloud agents, the Agents window, IDE, and CLI.
Allow remembrance.dev in the enterprise sandbox network allowlist for command-based fallbacks.
Cursor's current public enterprise documentation does not define a managed per-MCP-tool allowlist. Keep normal tool approvals enabled and do not claim an undocumented control exists.
When submit_private_lesson_candidate first appears, approve that individual tool for future invocations instead of enabling broad Auto-run.
Cloud-agent hook support can differ from local plugin support; verify query, invocation, feedback, and contribution receipts on every enabled surface.
On each surface, call get_connection_status and confirm organization scope before repository-derived contribution work; an installed plugin is not proof that its MCP credential is active.

Cursor team plugin install modes Cursor Team MCP distribution Cursor sandbox network controls Cursor cloud-agent hooks

VS CodeEnable the Preview agent-plugin and hook surfaces, deploy the reviewed Remembrance source, and confirm that enterprise MCP policy permits its bundled local server.Open

Enable the documented ChatPluginsEnabled and ChatHooks policies. A disabled plugin or hook policy removes the corresponding Remembrance lifecycle behavior.
Install the reviewed source with Chat: Install Plugin From Source using https://github.com/dreamarkinc/remembrance-skills, or distribute the same source through the organization's approved plugin process.
If ChatAllowedMcpServers or ChatDeniedMcpServers is configured, verify that the packaged remembrance MCP server is allowed and not denied. Deny rules take precedence.
Keep normal tool approvals enabled. Do not turn on global auto-approval to make Remembrance work.
Use Chat: Manage Tool Approval to pre-approve only submit_private_lesson_candidate; leave broad global auto-approval disabled.
Permit the local MCP process to reach remembrance.dev through the organization's endpoint security and egress controls. VS Code's agent-tool network filter is not documented as a policy for plugin subprocess traffic.
Provision the organization key through the mode-0600 shared Remembrance config or an approved process secret, restart VS Code, and call get_connection_status before private contribution work.

Copy VS Code policy.json additions

{
  "ChatPluginsEnabled": true,
  "ChatHooks": true
}

VS Code agent plugins VS Code enterprise policies VS Code MCP servers

OpenCodeDeploy the version-matched npm plugin and local MCP config, then use exact OpenCode tool permissions to keep reads low-friction and writes reviewable.Open

Run npx -y @remembrance-ai/opencode-plugin setup, or deploy the equivalent version-pinned opencode.json through the organization's normal configuration-management system.
OpenCode Enterprise supports centrally managed configuration. Self-managed teams can instead publish organization MCP defaults through .well-known/opencode and distribute the version-pinned plugin configuration with their normal configuration-management system.
Organization MCP defaults can be overridden by local configuration. Protect and validate managed opencode.json files, and prevent project config from disabling or duplicating the reviewed Remembrance integration.
Allow the exact read tools below and keep feedback and private mutations on ask. Optional public/shared contribution tools stay denied unless the organization approves them separately.
Set remembrance_submit_private_lesson_candidate to allow so the narrow private lesson lane stays hands-off without approving other mutations.
Permit the local MCP process to reach remembrance.dev through the organization's egress controls; do not broaden unrelated shell or web permissions.
Provision the organization key through the mode-0600 shared Remembrance config or an approved process secret, restart OpenCode, and call get_connection_status before private contribution work.

Copy OpenCode permission additions

{
  "$schema": "https://opencode.ai/config.json",
  "permission": {
    "remembrance_get_connection_status": "allow",
    "remembrance_run_connection_doctor": "allow",
    "remembrance_query_skills": "allow",
    "remembrance_list_skills": "allow",
    "remembrance_invoke_skill": "allow",
    "remembrance_get_effective_preferences": "allow",
    "remembrance_get_private_lesson_policy": "allow",
    "remembrance_get_skill": "allow",
    "remembrance_get_resource": "allow",
    "remembrance_get_value_proof": "allow",
    "remembrance_submit_query_feedback": "ask",
    "remembrance_submit_feedback": "ask",
    "remembrance_submit_preference_compatibility_feedback": "ask",
    "remembrance_report_task_outcome": "ask",
    "remembrance_submit_remembrance": "ask",
    "remembrance_propose_private_skill": "ask",
    "remembrance_submit_suggestion": "ask",
    "remembrance_record_preference": "ask",
    "remembrance_link_current_installation": "ask",
    "remembrance_submit_private_lesson_candidate": "allow",
    "remembrance_propose_skill_idea": "deny",
    "remembrance_submit_resource": "deny",
    "remembrance_submit_resource_review": "deny",
    "remembrance_request_attestation_challenge": "deny",
    "remembrance_register_agent_key": "deny"
  }
}

OpenCode plugins OpenCode tool permissions OpenCode MCP servers OpenCode Enterprise

Other MCP clientsRegister the exact hosted endpoint or bundled stdio command and expose only the approved organization tools supported by that client.Open

Allow only https://remembrance.dev/api/mcp for hosted MCP, or the exact @remembrance-ai/mcp-server command for stdio.
Expose the recommended organization tool set below; keep optional dual-scope/public, resource, and identity tools disabled unless explicitly approved.
Keep normal confirmation behavior for non-read-only tools unless the organization has explicitly approved unattended private contribution.
When supported, persist approval only for submit_private_lesson_candidate; preparation and queue management remain local-only.
If the host cannot express server or tool policy, rely on its existing destination approval and use the zero-network handoff when export is denied.
Call get_connection_status and confirm organization scope before any private contribution; never infer scope from an environment variable or an unrelated anonymous probe.

Copy recommended organization tool allowlist

{
  "endpoint": "https://remembrance.dev/api/mcp",
  "auth_header": {
    "X-Remembrance-API-Key": "${REMEMBRANCE_API_KEY}"
  },
  "enabled_tools": [
    "get_connection_status",
    "run_connection_doctor",
    "query_skills",
    "list_skills",
    "invoke_skill",
    "get_effective_preferences",
    "get_private_lesson_policy",
    "get_skill",
    "get_resource",
    "get_value_proof",
    "submit_query_feedback",
    "submit_feedback",
    "submit_preference_compatibility_feedback",
    "report_task_outcome",
    "submit_remembrance",
    "propose_private_skill",
    "submit_suggestion",
    "record_preference",
    "link_current_installation",
    "submit_private_lesson_candidate"
  ],
  "optional_shared_tools": [
    "propose_skill_idea",
    "submit_resource",
    "submit_resource_review",
    "request_attestation_challenge",
    "register_agent_key"
  ]
}

REMEMBRANCE_SUBMISSION_PAYLOAD { "schema_version": "0.1", "type": "skill_use", "task": {"domain": "web-ui-qa", "summary": "Redacted summary", "privacy": "redacted_public"}, "outcome": {"success": true, "usefulness_rating": 5, "confidence": 0.86}, "lesson": "What future agents should remember." }