RecruitMCP

Sourcing and outreach for recruiters, inside your AI agent

Your recruiting desk, inside Claude.

RecruitMCP gives Claude, Grok Bot or any MCP agent two things it can't do on its own: find candidates for a role without paying twice, and reach out at a pace that keeps your account safe. You ask in plain words. It shows the cost. Nothing runs until you say yes.

Set up RecruitMCP from https://recruitmcp.com/mcp.md

Paste the line into your agent. It reads the setup guide and connects itself.

Read the docsSee pricing1,250 free credits, no card

Claude · RecruitMCP connectedexample data
  1. You: Use RecruitMCP. Here's the job description for a Senior Backend Engineer in Ghent. Find 40 people within 30 km, 5+ years, Go or Kotlin. Skip anyone we've already contacted.
  2. Tool call: start_search { job_description: "Senior Backend Engineer, Ghent…", target: 40 }
  3. Check: ✓ Search profile drafted — 3 titles, 2 nearby titles, Ghent + 30 km, 5+ years
  4. Check: ✓ Plan built — 3 steps, no overlap between steps
  5. Waiting for you: ⏸ Estimate up to 120 credits. Run it?
  6. You: Yes.
  7. Tool call: start_search { search_id: "pln_93aD", confirm: true }
  8. Tool call: get_search_results { run_id: "run_5Tz8", wait_seconds: 25 }
  9. Check: ✓ 40 profiles found · 9 skipped (already contacted) · 120 credits spent
free credits, no card
1,250
credit per lead returned
1
credit per live profile
2
per outreach account / month
€49
invites / day, hard cap
20
API rate limit
120/min

// read from the same config the API enforces

How it works

You give it. It does the work. You decide.

  1. 01

    You give it

    A job description, a few lines about the role, or a list you already have.

  2. 02

    RecruitMCP does

    Turns it into a search plan, prices it, runs it after your OK, and hands back factual profiles. For outreach, it schedules every step inside safe limits and stops when someone replies.

  3. 03

    You decide

    Approve each step, or let a run go end to end within the budget you set.

Proof

One prompt, a shortlist.

The run in the terminal above ends in rows like these. Every value is a fact from the profile.

Claude · RecruitMCP connected · get_search_resultsexample data
Example candidates returned for a Senior Backend Engineer in Ghent
NameCurrent roleDistanceExperienceIn current role
A. De SmetSenior Backend Engineer, Acme Software6 km8 yrs3 yrs
J. PeetersSoftware Engineer (Go), Acme Logistics22 km6 yrs1 yr
M. ClaesPlatform Engineer, Acme Cloud28 km11 yrs5 yrs

Example data. Every row is a fact from the profile; RecruitMCP doesn't rank or judge — your agent does that, in your own words.

Module 01 · Sourcing

Find candidates in your agent.

Job text in, search plan out, factual profiles back. Public professional profiles, fetched live. You never pay twice for the same person.

SourcingDocs

  • The cost comes first.

    Every paid step returns an estimate. Your agent asks before it spends.

  • No double payment.

    Search steps never overlap, runs resume where they stopped, and people you've already seen are skipped before they cost anything.

Module 02 · Outreach

The missing piece.

Your agent decides who and what. RecruitMCP decides when — and whether it's safe on LinkedIn.

OutreachDocs

  • A human pace on LinkedIn.

    20 invites a day, 100 a week by default, a 14-day warm-up, working hours only.

  • Stops on reply.

    Before every follow-up, the conversation is checked again. A reply stops the sequence.

Why

Why the other ways stop working.

What you have tried, what breaks, and what RecruitMCP does instead
What you've triedWhat breaksWhat RecruitMCP does
Asking Claude to "find candidates"It has no live data and makes up namesReal profiles from a live search, with the facts behind each row
A sourcing tool with a fixed UIYour agent can't drive it; filters don't match how you briefPlain-language brief → editable plan your agent controls
A people APIYou pay for overlapping queries and the same people againNon-overlapping steps, a seen register, resumable runs
An automation tool for LinkedInBursts, identical daily counts, follow-ups after someone repliedSpread actions, daily jitter, warm-up, reply check before each send
Letting an agent send messagesNo brakesPreview, confirm, hard limits, auto-pause

Architecture

Your agent on top. Two modules underneath.

One server over MCP or REST. Both modules share the same brakes: an estimate before any spend, an explicit confirm, idempotency keys, and one audit log.

Your agent calls RecruitMCP over MCP or REST. RecruitMCP has two modules. Sourcing goes from intent to search plan to run to profiles, billed in credits, and reads public professional profiles. Outreach goes from account to sequence to enrollment to inbox behind a safety layer, billed per connected account, and sends from your own account, hosted sign-in. Both share estimate then confirm, idempotency keys, the credit ledger, the suppression list, and the audit log.

Plays

Plays to copy.

Paste any of these into your agent once RecruitMCP is connected.

Worked use cases

  • Plan first

    “Use RecruitMCP. Turn this job description into a search plan and tell me what it costs before running it.”

  • Only what's new

    “Re-run last week's search for the data engineer role. Only pay for what's new.”

  • Contacts for a few

    “Get personal email and mobile for the 8 people I starred. Show me the cost first.”

  • Sequence preview

    “Write a 3-step LinkedIn sequence for these 10 people, show me every message and when it would go out, then wait.”

  • Reply triage

    “Which of my outreach got a reply today? Draft an answer for each, don't send.”

Prefer code

The same steps over REST.

Every paid call accepts dry_run=true and an Idempotency-Key.

# Free: plan from the job text and read the estimate
curl -X POST "https://api.recruitmcp.com/v1/runs?dry_run=true" \
  -H "Authorization: Bearer $RECRUITMCP_KEY" \
  -H "Content-Type: application/json" \
  -d '{"job_description":"Senior backend engineer in Ghent, Go, 5+ years","target":40}'

# Spends credits: start the run once the estimate is OK
curl -X POST https://api.recruitmcp.com/v1/runs \
  -H "Authorization: Bearer $RECRUITMCP_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: gent-backend-01" \
  -d '{"plan_id":"pln_93aD","budget":{"max_credits":30}}'

Pricing

Pay for what runs, not for seats.

1,250 free credits to start. A returned lead is 1 credit, a profile 2. Planning, lookups, dedup and checks are free. Outreach is €49 per connected account per month, with no per-message fee — we don't want you to send more than is safe.

See pricing

Questions

Which agents work?

Claude (Desktop, claude.ai, Code), Grok Bot, and any client that speaks MCP. Or use the REST API.

Does it decide who's a good fit?

No. It returns facts. Your agent and you decide.

Will it send messages without me?

Only after you approve them. Replies always need your OK.

See the full FAQ

Bring one role and see what comes back.

Connect RecruitMCP, paste a job description, and read the plan and the price before anything runs.

Get started freeRead the docs