Use any MCP server
Add a remote MCP server and its tools become your agent's tools.
The Model Context Protocol (MCP) lets any service offer tools to AI agents. Add a remote MCP server as a connector and its tools become your agent's tools, with no code on your side.
Add a server
- In Connectors, choose MCP server.
- Paste its https address (the Streamable HTTP endpoint, often ending in
/mcp) and, if it needs one, a bearer token. - We connect, list its tools and show them on the card. Check fetches the list again after the server changes.
Build your own
Any MCP server that speaks Streamable HTTP works. A tiny one with the official TypeScript SDK:
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/streamableHttp.js";
import express from "express";
import { z } from "zod";
const server = new McpServer({ name: "orders", version: "1.0.0" });
server.tool("order_status", { id: z.string() }, async ({ id }) => ({
content: [{ type: "text", text: `Order ${id} shipped yesterday.` }],
}));
const app = express();
app.use(express.json());
app.post("/mcp", async (req, res) => {
const transport = new StreamableHTTPServerTransport({ sessionIdGenerator: undefined });
await server.connect(transport);
await transport.handleRequest(req, res, req.body);
});
app.listen(3000);Good to know
- Up to 40 tools of a server are offered to an agent; keep descriptions short and clear, they are what the model reads.
- Text answers work best; images and audio are mentioned, not shown.
- Each tool call costs 2 credits.