@weaveintel/mcp-server

Expose your own tools, resources, and prompts over the open Model Context Protocol.

Why it exists

You've built tools you'd like other AI apps to use — a desktop assistant, an editor, a colleague's agent. The Model Context Protocol is the USB-C port for AI: one standard socket everyone already knows how to plug into. This package turns your tools into that socket. You describe what you offer once, pick a transport, and any MCP-speaking client can discover and call it — no bespoke API, no custom SDK for each consumer.

When to reach for it

Reach for it when you want to publish capabilities for the outside world to consume. If instead you want your agent to use someone else's MCP server, that's the opposite direction — use @weaveintel/mcp-client. And if you don't need a live transport at all — say you're embedding MCP inside another HTTP route — drive the transport-free handleMcpMessage core directly.

How to use it

import {
  weaveMCPServer,
  createMCPStdioServerTransport,
  mcpText,
} from '@weaveintel/mcp-server';

const server = weaveMCPServer({ name: 'my-tools', version: '1.0.0' });

server.addTool(
  { name: 'greet', description: 'Say hello', inputSchema: { type: 'object' } },
  async (_ctx, _args) => mcpText('Hello from my server!'),
);

await server.connect(createMCPStdioServerTransport());

What's in the box

| Export | What it does | | --- | --- | | weaveMCPServer(config, opts?) | The server: addTool, add resources/prompts, connect | | createMCPStdioServerTransport() | Serve over stdio (e.g. to a desktop client subprocess) | | createMCPStreamableHttpServerTransport() | Serve over streamable HTTP | | weaveRealMCPTransport(opts) | Batteries-included real HTTP transport server | | handleMcpMessage, mcpText, MCP_PROTOCOL_VERSION | Transport-free JSON-RPC core + helpers |

License

MIT.