MCP Server

The BrainOS MCP server gives Claude Desktop and Claude Code direct access to your AI Workers via the Model Context Protocol.

Install

terminalbash

pnpm add @brainos/mcp-server

Claude Desktop Configuration

Add the following to your Claude Desktop configuration file:

claude_desktop_config.jsonjson

{
  "mcpServers": {
    "brain-os": {
      "command": "npx",
      "args": ["@brainos/mcp-server"],
      "env": {
        "SUPABASE_URL": "https://your-project.supabase.co",
        "SUPABASE_ANON_KEY": "your-anon-key",
        "BRAINOS_ORG_ID": "org_123"
      }
    }
  }
}

Available Tools

The MCP server exposes 5 tools that Claude can use during conversations:

brainos_query

Query an AI Worker with Brain IQ routing. Claude sends a question and receives an answer with confidence score and reasoning trace.

brainos_query({ query: "Why is churn rising?", domain: "cs" })

brainos_ingest

Send signals to the AI Worker brain. Claude can ingest observations and outcomes during a conversation.

brainos_ingest({ signals: [{ source_domain: "cs", signal_type: "escalation", signal_value: 1 }] })

brainos_knowledge

Read federated knowledge edges. Returns learned knowledge for a domain or AI Worker.

brainos_knowledge({ domain: "finance", minConfidence: 0.7 })

brainos_webhook

Forward webhooks. Claude can relay external events to BrainOS for processing.

brainos_webhook({ source: "stripe", payload: { ... } })

brainos_sleep

Trigger brain maintenance. Run RL consolidation, federated sync, or sleep cycle.

brainos_sleep({ jobs: ["rl-consolidate", "federated-sync"] })

Knowledge Federation

The MCP server automatically federates between the Core Memory (universal knowledge) and your Org Memory (proprietary knowledge). Your org memory always takes priority. When Claude asks a question, both memories contribute to the answer, with clear labelling of which insights come from universal vs. organisational data.