MCP Server Setup
The KnowledgePulse MCP server (@knowledgepulse/mcp v1.1.0) exposes the KnowledgePulse protocol as a set of Model Context Protocol tools that any MCP-compatible AI client can call.
Transport
The server uses Streamable HTTP transport:
| Endpoint | Method | Description |
|---|---|---|
/mcp | POST | MCP tool invocation (Streamable HTTP) |
/health | GET | Health check |
A successful health check returns:
{
"status": "ok",
"name": "knowledgepulse-mcp",
"version": "1.1.0"
}
Dual-Mode Operation
The MCP server can run in two modes depending on whether the KP_REGISTRY_URL environment variable is set.
Standalone Mode (default)
In standalone mode the server uses its own in-memory stores. This is the simplest way to get started and is well-suited for local development and testing.
bun run packages/mcp-server/src/index.ts
The server starts on port 3001 by default. No external services are required.
Proxy Mode
In proxy mode the server forwards all requests to a running KnowledgePulse registry instance. Set KP_REGISTRY_URL to enable proxy mode, and optionally provide KP_API_KEY for authenticated endpoints.
KP_REGISTRY_URL=http://localhost:8080 KP_API_KEY=kp_abc123 \
bun run packages/mcp-server/src/index.ts
In this mode the MCP server acts as a thin bridge: it translates MCP tool calls into registry REST API requests and returns the results to the client.
Environment Variables
| Variable | Description | Default |
|---|---|---|
KP_MCP_PORT | Port the MCP server listens on | 3001 |
KP_REGISTRY_URL | Registry URL for proxy mode. When unset the server runs in standalone mode. | (unset) |
KP_API_KEY | API key sent with authenticated requests to the registry in proxy mode. | (unset) |
Integration with AI Frameworks
The MCP server works with any MCP-compatible client over Streamable HTTP transport. Examples include:
- Claude Desktop -- add the server URL to your MCP configuration.
- LangGraph -- use the MCP tool adapter to connect to the server.
- CrewAI -- register the server as an MCP tool provider.
- AutoGen -- connect agents to the server via the MCP client SDK.
Point your client at http://localhost:3001/mcp (or whichever host and port you configured) and the six KnowledgePulse tools will be available for your agents to call. See MCP Tools for a full reference.