person-search-mcp

An MCP server that researches a real person from public web sources and returns a cited briefing — drop it into any MCP-capable agent (Claude Desktop, Cursor, Cline, Windsurf, OpenClaw, …) and your agent gains a search_person tool.

It's a thin, generic wrapper over the public PersonSearch "Deep Search" agent. Give it a name and a public profile link; it runs federated web research and returns a structured, source-cited markdown report (current role, background, online presence, notable quotes). No API key required.

You:    Research Paul Graham (https://x.com/paulg) before my call.
Agent:  → calls search_person → returns a cited briefing with numbered sources.

The tool

search_person

| Argument | Required | Description | | --- | --- | --- | | name | yes | Full name, e.g. "Paul Graham". | | profile_url | yes | A public profile link (LinkedIn, personal site, GitHub, X, …). Anchors the research to the right individual. | | context | no | Extra disambiguating context, e.g. "Founder, Acme". | | force_refresh | no | Re-run from scratch, bypassing any cached result. |

Returns the briefing markdown followed by a numbered Sources list. For people with little public presence it reports gaps honestly instead of inventing facts (and flags low evidence quality).

Quick start

Requires Node.js ≥ 18.

Run it in an MCP client

Add the server to your client's MCP config. Today (no install needed), run it straight from GitHub:

{
  "mcpServers": {
    "person-search": {
      "command": "npx",
      "args": ["-y", "github:DhanushKenkiri/PersonSearchAgent-MCP"]
    }
  }
}

Once published to npm, swap the args to ["-y", "person-search-mcp"].

Config file locations:

• Claude Desktop — claude_desktop_config.json (mcpServers key)
• Cursor — .cursor/mcp.json
• Cline / Windsurf / others — their MCP settings JSON (same shape)

Or run from a clone

git clone https://github.com/DhanushKenkiri/PersonSearchAgent-MCP.git
cd PersonSearchAgent-MCP
npm install        # builds automatically via the prepare script
node dist/index.js # starts the MCP server on stdio

Then point your client's command/args at node + the absolute path to dist/index.js.

Try it without an MCP client (CLI)

node dist/index.js search "Paul Graham" "https://x.com/paulg" "Y Combinator"
# or, from anywhere, with no clone:
npx -y github:DhanushKenkiri/PersonSearchAgent-MCP search "Paul Graham" "https://x.com/paulg"

OpenClaw

This works with OpenClaw agents too — see openclaw/README.md for both paths:

• MCP — register this server through OpenClaw's built-in mcporter skill, or
• Native skill — drop the ready-made person-search skill into ~/.openclaw/skills/.

Configuration

| Env var | Default | Purpose | | --- | --- | --- | | PERSON_SEARCH_API_URL | https://mvp.actualte.tech | Base URL of the research backend. Point it at your own deployment to self-host. | | PERSON_SEARCH_TIMEOUT_MS | 180000 | Per-request timeout (deep research can take 1–2 minutes). |

The default backend is the hosted, public PersonSearch agent, so the server works out of the box. It calls a single public endpoint — POST /api/research — and adds no auth, storage, or state of its own.

Develop

npm install     # install deps + build
npm run build   # compile src/ -> dist/
npm test        # build + run the MCP protocol tests (offline, deterministic)

The test suite (test/mcp.test.mjs) launches the built server over stdio and drives it with the official MCP SDK client against a local mock backend — verifying the initialize handshake, tools/list, a successful tools/call, and the error path.

Publish (maintainer)

npm login
npm publish --access public

After publishing, anyone can run it with npx -y person-search-mcp.

License

MIT © Dhanush Kenkiri