The playbook, as an API

Twenty years of sourcing craft behind one endpoint: send a role description, get back boolean and x-ray search strings written the way a working recruiter writes them. For AI agents, developers — and recruiters who just want it inside Claude or ChatGPT.

Three ways in, from zero-setup to pay-per-call:

Recruiters: add it to Claude or ChatGPT (30 seconds, free)

No wallet, no key, no install. Add the connector once and then just ask your assistant for search strings — it does the rest.

Connector URL: https://sourcerdesk.com/mcp

Claude: Settings → Connectors → Add custom connector → paste the URL.

ChatGPT: Settings → Apps & Connectors → Developer mode → Create → paste the URL (no authentication).

Then: "Search strings for a payroll manager, multi-state, in Philadelphia — LinkedIn and open web." Generation takes a minute or two. Free connector calls come from a shared daily pool; when it runs dry, the tool points you at the API below or the Chrome extension — the full desk, unlimited.

Developers and agents: the endpoint

Try it free right now (3 calls/day/IP, responses take 60–180 seconds):

curl -X POST https://sourcerdesk.com/api/v1/boolean \
  -H "Content-Type: application/json" \
  -d '{"role_description": "corporate paralegal, M&A closings",
       "location": "Chicago"}'

The response carries expanded title variants, practitioner "shibboleth" skill terms, ready-to-paste LinkedIn x-ray and open-web resume strings, and iteration advice. Full schema and examples: the documentation repo.

Past the free tier: ~2¢ a call, three rails

RailNetworkPrice per call
Nano (XNO) — feelessnano:mainnet0.05 XNO ≈ $0.02
USDCBase$0.025
USDCPolygon$0.025

Payment is the open x402 standard (HTTP 402): the endpoint quotes, your client pays, the receipt settles on-chain. The feeless Nano rail is deliberately the cheapest — no gas, no minimum, the metered amount is the price. Agents on Nano: pip install feeless402. Standard USDC x402 clients work as-is, gas-free. No account with us, ever — the first request quotes, the second one is paid.

Why it beats asking a model directly

Before launch, the engine had to pass a pre-registered blind test: ten roles, engine vs. the same model with a generic prompt, sides shuffled, answer key sealed, graded by the recruiter whose playbook it encodes. The engine won 10 out of 10. The full methodology — including where its evidence is thinner than the headline — is published in EVAL.md. The playbook itself stays private; the way it was tested doesn't.

Links

Documentation repo · machine docs at /api and /llms.txt · MCP registry: com.sourcerdesk/sourcers-desk · questions: support@sourcerdesk.com