@firetrace/mcp

Model Context Protocol server for FireTrace. Lets an AI agent list, inspect, record, and delete LLM traces through a scoped project API key.

stdio bridge

FIRETRACE_ENDPOINT=https://your-deployment.vercel.app \
FIRETRACE_API_KEY=ft_live_... \
npx -y @firetrace/mcp

The bridge validates the key against GET /api/v1/key, then exposes only the tools the key's scopes allow (traces:read, traces:write, traces:delete). Diagnostics go to stderr; stdout is the MCP stream.

Most clients can instead talk to the deployment directly at POST /api/mcp with the same bearer key. See docs/mcp.md in the repository for client configuration and the tool reference.

Library use

import { createFireTraceMcpServer, HttpBackend } from "@firetrace/mcp";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const backend = new HttpBackend({
  endpoint: "https://your-deployment.vercel.app",
  apiKey: "ft_live_...",
});
await backend.init();
const server = createFireTraceMcpServer(backend);
await server.connect(new StdioServerTransport());

Implement TraceBackend to put the same tools over any other store.

MIT.