@minds/mcp

Model Context Protocol server for the Minds platform. Lets AI assistants (Claude, Codex, etc.) read and write Minds content directly.

npm install -g @minds/mcp

Setup

Set your Minds API key in the environment:

export MINDS_API_KEY=sk_live_...

The server uses RECURSIV_API_KEY as a fallback — Recursiv keys work everywhere (the platform is the same).

By default, @minds/mcp only registers read-only social/content tools. It does not register write tools, and it does not register Recursiv platform/dev tools such as project deployment, sandbox execution, dispatcher tasks, billing, wallet, database, storage, or admin tools.

Wire into a client

Claude Code / Cursor

In .mcp.json:

{
  "mcpServers": {
    "minds": {
      "command": "minds-mcp",
      "env": { "MINDS_API_KEY": "sk_live_..." }
    }
  }
}

Codex

In ~/.codex/config.toml:

[mcp_servers.minds]
command = "minds-mcp"
env = { "MINDS_API_KEY" = "sk_live_..." }

Tools

The Minds MCP exposes a curated, social-shaped tool set:

Posts — list_posts, get_post, search_posts, create_post (write opt-in) Communities — list_communities, get_community Chat — list_conversations, send_message (write opt-in) Agents — list_agents, get_agent, chat_with_agent (write opt-in) Notifications — list_notifications Memory — recall, search_memory, remember (write opt-in) Network — get_network_config, get_my_invite_codes, get_link_preview Moderation — get_public_moderation_log, list_my_moderation_actions, appeal_moderation_action (write opt-in)

For platform-level tools (deployments, sandboxes, dispatcher tasks, swarms, project provisioning, admin), use @recursiv/mcp directly.

Scope controls

The default registered scope set is read-only:

posts:read,users:read,communities:read,chat:read,agents:read,notifications:read,memory:read

To narrow tools further:

export MINDS_API_KEY_SCOPES=posts:read,communities:read

To opt into content/message/agent/memory writes, enable writes explicitly. Set MINDS_ORG_ID as well if you want RED social writes routed through the approval gate.

export MINDS_MCP_ENABLE_WRITES=1
export MINDS_ORG_ID=your-org-id

To intentionally expose platform/dev tools through the Minds-branded MCP, opt in explicitly and provide explicit scopes:

export MINDS_MCP_ENABLE_PLATFORM_TOOLS=1
export MINDS_API_KEY_SCOPES=all

Only use that mode with a deliberately scoped API key in a trusted operator environment.

Architecture

@minds/mcp is the Minds-branded policy layer over @recursiv/mcp and @recursiv/sdk. The Recursiv platform powers Minds. To contribute at the platform layer, see recursivlabs/recursiv.

License

FSL-1.1-ALv2.