When you connect an agent to the Get on Board MCP Server, the fastest way to get value is to hand it a complete recipe: a goal, the tools to use, and where to stop. Below are six sourcing recipes you can paste into Claude (Desktop or Code), ChatGPT, Codex, or Cursor once your connection is set up. All of them use the server’s read-only tools — search_talent, talent_profile, list_jobs, job_details, and whoami — and every one ends the same way: a shortlist of Get on Board profile links you open in the web app to unlock and contact.
This article is for recruiters with the MCP server already connected. If you have not set it up yet, start with What is the Get on Board MCP Server? or the setup guide for Claude, ChatGPT, Cursor, or Codex.
How these recipes work
Every recipe follows the same shape:
- State the goal in plain language. The
search_talenttool takes a natural-language brief — role, stack, seniority, country, English level — not filter IDs. Write the search the way you would explain it to a colleague. - Let the agent post-filter. Results come with country, seniority, years of experience, English level, skills, and a recruiter-facing summary. Ask the agent to filter and rank on those fields instead of running one giant query.
- End at links. Contact details never travel through MCP. A finished recipe produces profile URLs; unlocking happens in the web app, with credits.
Profile reads are capped per hour, so recipes ask for deep reads (talent_profile) only on the candidates that survive the first cut.
Recipe 1: Shortlist for a job you already posted
The highest-leverage recipe: the job description you wrote becomes the search brief, with no re-typing. Works in any connected client.
List my open jobs, then pull the full details of the [role name] one. Using its requirements, search Talent Database for matching candidates. Post-filter to [country/countries] and [English level], read the full profile of the 5 strongest matches, and give me a ranked shortlist with each candidate’s profile link and a two-line reason.
The agent chains list_jobs → job_details → search_talent → talent_profile, then hands you links.
Recipe 2: Source from a brief you have not posted yet
Useful for confidential roles or early pipeline building, before a job exists. Works in any connected client.
Search Talent Database for senior backend developers in Chile or Colombia with Elixir or Erlang experience and conversational English, open to remote work. Show me the top 10 with country, seniority, years of experience, and main skills, and flag which ones are already unlocked for my company.
Search results include a locked/unlocked flag, so the agent can tell you which candidates your company can already see in full.
Recipe 3: Compare finalists against a role
When you are down to a few candidates and want a structured read before deciding whom to unlock or move forward. Works in any connected client.
Here are 3 candidate ids from the earlier search: [ids]. Read each full profile and compare them against the requirements of my [role name] job. Build a table: candidate, strongest evidence for the role, gaps or open questions, and what I should verify in an interview. Include each profile link.
The agent combines talent_profile reads with job_details and does the side-by-side you would otherwise build by hand.
Recipe 4: Widen or narrow an unproductive search
Searches are conversational, so treat the first query as a draft. Works in any connected client.
That search returned too few strong results. Try two variations: one that drops the [specific framework] requirement and keeps the domain experience, and one that keeps the stack but adds neighboring countries. Tell me which variation produced better candidates and why.
Because the agent post-filters on the returned fields, you can relax one requirement at a time and see exactly what each constraint was costing you.
Recipe 5: Coverage check across all your open roles
A pipeline health report instead of a single search. Works in any connected client; roles with long lists read best in Claude Code, Codex, or Cursor, where output is easy to scroll and save.
List all my open jobs. For each one, run a quick Talent Database search based on its title and main requirements and report roughly how deep the candidate pool looks: strong, workable, or thin. Give me the list ordered from thinnest to deepest so I know where to put sourcing effort this week.
Recipe 6: Weekly sourcing digest
A repeatable brief you re-run on a schedule. Best in Claude Code, Codex, or Cursor, because those clients can save each week’s shortlist to a file and compare runs; in Claude Desktop or ChatGPT you can keep the previous list in the conversation instead.
Re-run my saved brief: [brief]. Compare against last week’s shortlist file and tell me who is new, who dropped out, and whether any new candidate beats my current top 3. Save today’s shortlist with date, ids, and profile links.
Access tokens last 2 hours, so a scheduled or spaced-out run will ask you to re-authorize — that is expected, not an error.
What no recipe can do
However you phrase a recipe, an agent connected through MCP cannot unlock contact details, invite candidates, apply to jobs, post jobs, or see anyone’s email, phone, CV, or social links. Those steps stay in the web app, where you decide and where credits are spent. The full boundary map is in What can an AI agent do (and not do) on Get on Board?
More from Talent database and sourcing
Sign in with Google
