@weaveintel/mcp-client

Connect to any external MCP server and hand its tools straight to your agent.

Why it exists

Lots of useful capabilities now live behind the Model Context Protocol — a filesystem server, a database server, a company's internal server. Each one is a little appliance with its own plug. This package is the extension cord: point it at a server, and every tool that server offers shows up in your agent's tool registry as if you'd written it yourself. You don't learn each server's wiring; you just plug in.

When to reach for it

Reach for it when you want your agent to use tools that some other MCP server provides. If instead you want to publish your own tools for others to consume, that's the other end of the cord — use @weaveintel/mcp-server.

How to use it

import {
  weaveMCPClient,
  weaveMCPTools,
  createMCPStreamableHttpTransport,
} from '@weaveintel/mcp-client';
import { weaveContext } from '@weaveintel/core';

const client = weaveMCPClient();
await client.connect(createMCPStreamableHttpTransport('https://example.com/mcp'));

const tools = await client.listTools();          // what the server offers
const registry = weaveMCPTools(client, tools);    // drop-in ToolRegistry for your agent

const ctx = weaveContext();
const result = await client.callTool(ctx, { name: tools[0].name, arguments: {} });
console.log(result.content);

What's in the box

| Export | What it does | | --- | --- | | weaveMCPClient() | The client: connect, listTools, callTool (plus resources & prompts) | | weaveMCPTools(client, tools) | Wraps discovered tools into a ToolRegistry your agent can call | | createMCPStreamableHttpTransport(endpoint, opts?) | Talk to a server over HTTP (with custom/dynamic headers) | | createMCPStdioClientTransport(params) | Talk to a server running as a local subprocess (stdio) |

License

MIT.