Metalogue
Docs

Expose

MCP server for exposing Metalogue to AI tools and agents.

Expose

Expose is the AI-native integration layer. Deploy Metalogue as an MCP (Model Context Protocol) server, enabling Claude, ChatGPT, and other AI tools to query your organizational knowledge.

Overview

graph LR
    subgraph "AI Tools"
        CLAUDE[Claude Desktop]
        CURSOR[Cursor]
        COPILOT[GitHub Copilot]
    end
    
    subgraph "Metalogue"
        MCP[MCP Server]
        API[API Layer]
        FCRS[FCRS Core]
    end
    
    CLAUDE -->|MCP| MCP
    CURSOR -->|MCP| MCP
    COPILOT -->|MCP| MCP
    
    MCP --> API
    API --> FCRS

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external data sources. Instead of pasting context into prompts, AI tools can:

  1. Discover available tools and resources
  2. Query data in real-time
  3. Execute actions on your behalf

Metalogue implements MCP to give AI tools access to your entire organizational memory.

Setting Up

1. Get Your MCP Endpoint

Your MCP endpoint is available at:

https://api.metalogue.xyz/v1/expose/mcp

Or self-host the MCP server:

# Clone and run
git clone https://github.com/metalogue/mcp-server
cd mcp-server
npm install
METALOGUE_API_KEY=mlo_xxx npm start

2. Configure Claude Desktop

Add to your Claude Desktop configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "metalogue": {
      "command": "node",
      "args": ["/path/to/mcp-server/index.js"],
      "env": {
        "METALOGUE_API_KEY": "mlo_your_api_key"
      }
    }
  }
}

Or use the hosted endpoint:

{
  "mcpServers": {
    "metalogue": {
      "url": "https://api.metalogue.xyz/v1/expose/mcp",
      "headers": {
        "Authorization": "Bearer mlo_your_api_key"
      }
    }
  }
}

3. Configure Cursor

In Cursor settings, add:

{
  "mcp": {
    "servers": {
      "metalogue": {
        "url": "https://api.metalogue.xyz/v1/expose/mcp",
        "apiKey": "mlo_your_api_key"
      }
    }
  }
}

Available Tools

The MCP server exposes these tools to AI assistants:

query_knowledge

Search across all connected sources:

// Tool signature
{
  name: "query_knowledge",
  description: "Search organizational knowledge across Slack, Notion, GitHub, etc.",
  parameters: {
    query: { type: "string", required: true },
    sources: { type: "array", items: "string" },
    limit: { type: "number", default: 5 }
  }
}

// Example usage by Claude
"Let me search for information about authentication decisions..."
[Calls query_knowledge with query="authentication architecture decision"]

get_document

Retrieve a specific document:

{
  name: "get_document",
  description: "Get full content of a specific document",
  parameters: {
    document_id: { type: "string", required: true }
  }
}

list_connectors

List available data sources:

{
  name: "list_connectors",
  description: "List all connected data sources",
  parameters: {}
}

execute_action

Perform actions on connected tools:

{
  name: "execute_action",
  description: "Execute an action on a connected tool (e.g., send Slack message)",
  parameters: {
    connector_id: { type: "string", required: true },
    intent: { type: "string", required: true },
    context: { type: "object" }
  }
}

Resources

The MCP server also exposes resources:

Connected Sources

{
  uri: "metalogue://connectors",
  name: "Connected Data Sources",
  description: "List of all connected tools and their sync status"
}

Recent Documents

{
  uri: "metalogue://recent",
  name: "Recent Documents",
  description: "Most recently synced documents across all sources"
}

API Endpoints

MCP WebSocket

wss://api.metalogue.xyz/v1/expose/mcp/ws

The MCP server communicates via WebSocket for real-time bidirectional messaging.

MCP HTTP (Fallback)

POST /v1/expose/mcp/invoke
Content-Type: application/json

{
  "tool": "query_knowledge",
  "parameters": {
    "query": "authentication decision",
    "limit": 5
  }
}

List Available Tools

GET /v1/expose/mcp/tools

Response:

{
  "tools": [
    {
      "name": "query_knowledge",
      "description": "Search organizational knowledge",
      "parameters": {
        "query": { "type": "string", "required": true },
        "sources": { "type": "array" },
        "limit": { "type": "number" }
      }
    }
  ]
}

List Resources

GET /v1/expose/mcp/resources

Example Conversation

Once configured, AI assistants can naturally query your knowledge:

User: "What did the team decide about the authentication architecture last month?"

Claude:

Let me search our organizational knowledge for this...

[Querying Metalogue...]

Based on the Slack discussions and Notion documentation I found:

The team decided on a centralized auth service using the singleton pattern on January 15th. Key points:

  • Single source of truth for auth state
  • JWT tokens with 24-hour expiry
  • Refresh tokens stored in Redis

Sources:

  • Slack #engineering (Jan 15)
  • Notion "Auth Architecture" page
  • GitHub PR #1234 discussion

Security

Authentication

All MCP connections require a valid API key:

  • WebSocket: Authorization header on connection
  • HTTP: Authorization: Bearer mlo_xxx header

Permission Scoping

MCP respects your API key's permissions:

PermissionMCP Capability
queryquery_knowledge, get_document
connector:readlist_connectors
connector:writeexecute_action

Audit Logging

All MCP interactions are logged:

{
  "event_type": "mcp.tool_invoked",
  "tool": "query_knowledge",
  "user_id": "...",
  "timestamp": "2026-01-20T02:00:00Z"
}

Self-Hosting

For enterprise deployments, self-host the MCP server:

# Docker
docker run -p 3333:3333 \
  -e METALOGUE_API_KEY=mlo_xxx \
  metalogue/mcp-server

# Node.js
git clone https://github.com/metalogue/mcp-server
cd mcp-server
npm install
METALOGUE_API_KEY=mlo_xxx npm start

Configuration

{
  "port": 3333,
  "metalogue_api_key": "mlo_xxx",
  "metalogue_base_url": "https://api.metalogue.xyz",
  "allowed_origins": ["https://your-domain.com"],
  "rate_limit": {
    "requests_per_minute": 60
  }
}

Troubleshooting

Connection Issues

# Test WebSocket connection
wscat -c wss://api.metalogue.xyz/v1/expose/mcp/ws \
  -H "Authorization: Bearer mlo_xxx"

Tool Not Found

Ensure your API key has the required permissions for the tool you're calling.

Rate Limiting

MCP requests count against your API rate limits. If rate limited, the tool returns an error.

Next Steps