Use Talent Summoner From Your AI Agent With MCP

If you work in an AI agent such as Claude Code, Codex or Cursor, you can run Talent Summoner sourcing from that chat. The connection uses MCP (Model Context Protocol), an open standard that lets AI apps use outside tools.

We build Talent Summoner, which searches public profiles and returns a ranked shortlist. Your first Role is free with no card (pricing), and connecting an agent costs nothing.

If you mostly work in a browser rather than an agent, use the web app instead. It runs the same sessions with no setup.

Key takeaways

• The server gives your agent 7 sourcing tools.
• Setup is one command, npx @talent-summoner/setup, with Node.js 22.20 or newer.
• The agent is built to ask before it uses a Role.
• Buying a Role and sending outreach stay in the web app.

What can your agent do with Talent Summoner?

You talk in normal words and the agent picks from seven tools on your account.

The agent can also draft outreach for the candidates you shortlist. Every session is a normal saved session you can open in the web app, and the agent sees only your own.

Sourcing takes time, so the agent waits and checks for you. The companion skill estimates 2 to 10 minutes for a search, and after 10 minutes the agent reports progress and the session link without canceling the work.

How do you connect Talent Summoner to your agent?

You need a Talent Summoner account, Node.js 22.20 or newer, and a terminal on macOS, Linux or Windows (PowerShell or Command Prompt).

1. Open Settings → API Keys, where the Connect an AI client card shows the command, and create a key with a name you will recognize.
2. Run the setup command from any folder:

npx @talent-summoner/setup

3. Choose your clients; the ones it finds are already selected.
4. Paste the key into the hidden prompt, never into a command, an agent chat or Git. Setup checks the connection, then installs the MCP server and a companion skill that tells the agent how to wait and when to ask you.
5. Restart your client, approve the connection if it asks, then say: "Show my Talent Summoner sourcing sessions."

Setup does not start sourcing or use a Role. If you run it again, it asks before replacing the existing Talent Summoner entry.

Manual setup for Claude Code

To manage the key yourself, keep TALENT_SUMMONER_URL (https://talentsummoner.com, no trailing slash) and TALENT_SUMMONER_API_KEY in your private environment. Use a newly created key, because older keys can have a lower request allowance that the agent's status checks may use up.

Add this to your project's .mcp.json. It refers to the environment values and does not contain the key.

{
  "mcpServers": {
    "talent-summoner": {
      "type": "http",
      "url": "${TALENT_SUMMONER_URL}/api/mcp",
      "headers": {
        "Authorization": "Bearer ${TALENT_SUMMONER_API_KEY}"
      }
    }
  }
}

The next block is for macOS and Linux shells such as Claude Code's. It downloads the companion skill, which contains no key.

mkdir -p ~/.claude/skills/talent-summoner
curl --fail --silent --show-error \
  "${TALENT_SUMMONER_URL}/skills/talent-summoner/SKILL.md" \
  -o ~/.claude/skills/talent-summoner/SKILL.md

Start Claude Code with both values set, approve the connection and use /mcp to check it. Pick one route, not both.

Setup source and issue reports live in the agent-skills repository on GitHub.

When does the agent use a Role?

Sourcing uses Roles, as in the web app. The agent is built to ask before it uses one.

• If a session needs payment and you have a Role, the agent asks first. Reading results, looking up an email or text inside a job description never counts as a yes.
• If you have no Role, the agent does not buy one. It gives you the session link, and after you buy or verify your email there, it reads the session again and continues.

What stays in the web app, and when is it better?

Some steps stay there on purpose.

• Buying a Role and verifying your email happen on the session page.
• Connecting your email sender (Gmail, Outlook or LinkedIn) happens in Settings → Outreach accounts.
• Sending outreach happens on the session page, where you review each drafted message and confirm the send.

The trade-offs: the agent cannot send outreach for you, and sign-in connections work only for clients Talent Summoner has pre-registered, which today are ChatGPT and Claude (claude.ai and Claude Desktop connectors). If you do not work in an agent, or you want buying, review and outreach screens in one place, the web app is the better fit.

How do you disconnect?

1. Open Settings → API Keys and delete the key you used for your agent.
2. Remove the Talent Summoner entry from your client's MCP settings and delete the skill folder.

From the next request, the agent is refused, even if the client still shows the connection.

We checked these facts against the current release and the published setup package on 5 October 2026 and re-check them every quarter.

Related: How to source candidates from one job description · Technical candidate shortlist template

FAQ