Pin MCP server

The recruiting MCP for Claude, Codex, and everything else.

Q: What OAuth scopes does the Pin MCP request?

One: `mcp`. It grants tool access inside your Pin workspace and nothing else. Billing, team admin, and org settings are not reachable through the MCP under any circumstances.

Q: What tools does the server expose?

Ten, grouped by phase. Setup: `list_jobs`, `scrape_job_description`, `create_job`. Review: `get_candidates`, `accept_candidate`, `reject_candidate`. Tuning: `modify_search`, `recalibrate_search`. Memory: `read_job_memory`, `update_job_memory`.

Q: Can the assistant send outreach without my approval?

No. Outreach is only queued when `accept_candidate` fires, and that tool call is shown to you by your client before it runs. There is no draft-and-send-silently path. You can also revoke the token at any time and in-flight calls fail immediately with 401.

Q: How does the memory work?

Each job has a per-workspace memory slot. `update_job_memory` persists a running summary — what you accept, what you decline, patterns the assistant spotted, and its current confidence. `read_job_memory` pre-loads that summary on the next session. The assistant uses it to pre-judge candidates the way you would.

Run Pin from the assistant you already work in. Start a search, review candidates with it, let it learn your taste, and hand it the rest when you're ready. Same Pin, same 850M+ profiles, in plain language.

Connect to Claude Connect to Codex Other clients

Included on every paid Pin plan · See pricing

profiles searchable: 850M+; profiles searchable
MCP tools exposed: 10; MCP tools exposed
from install to first call: 60s; from install to first call
auth — no API keys: OAuth 2.1; auth — no API keys

In your chat window

Brief and match, without leaving your assistant.

Real prompts, real Pin tool calls. The assistant starts the search, explains why each person fits, and waits for your call. Accepted candidates go straight into your Pin outreach sequence.

Illustrative

Here's the Greenhouse link for our Senior iOS role. Scrape it, start the search, and pull me the first batch of 10.

Pin scrape_job_description done

url: boards.greenhouse.io/acme/…

Senior iOS Engineer · New York, NY

Pin create_job done

title: Senior iOS Engineer · company: Acme

Search created · sourcing in the background

Pin get_candidates done

limit: 10

10 unreviewed candidates

Here's the first batch. Maya and Jamal both have years of production SwiftUI. Want to go through them one at a time?

From 850M+ profiles

get_candidates · 10 results

Maya Chen

Brooklyn, NY · 9 yrs

SwiftSwiftUI Yes
Jamal Okafor

Manhattan, NY · 11 yrs

SwiftUIKit Yes
Priya Shankar

Jersey City, NJ · 8 yrs

SwiftUIAccessibility Maybe

For each candidate in this batch: 2–3 sentence fit summary, predict accept or reject, then ask me. Skip the preamble.

Pin get_candidates done

job: Senior iOS Engineer

Batch loaded with fit hints

Maya Chen: 9 years of iOS, ships SwiftUI daily, Brooklyn-based. I'd predict accept. Accept or pass?

Maya Chen

Brooklyn, NY

Reject Accept

Production SwiftUI? Yes, ships it daily.

NYC or remote? Yes, Brooklyn.

You: “Accept Maya — she's exactly the shape.”

accept_candidateconfirmed by you · outreach queued

You've noticed I've rejected 5 in a row for the same reason. Lock in what I prefer, recalibrate the ranker, and go again.

Pin update_job_memory done

note: Never below 5 yrs of production Swift

Memory saved for this job

Pin recalibrate_search done

signal: 5 rejects · seniority

Ranker updated

Pin get_candidates done

limit: 10

Fresh, better-matching batch

Saved. I'll skip anyone under five years of production Swift from now on, here and in future sessions. New batch is ready.

Remembered next session

Job memory · Senior iOS Engineer

Never below 5 yrs of production Swift
De-rank bootcamp-only backgrounds
NYC metro or remote only

Loaded with read_job_memory at the start of the next session.

Definition

What is the Pin MCP?

The Pin MCP is a Model Context Protocol server from Pin (pin.com) that lets AI assistants run a recruiting workflow on your behalf — kicking off searches, reviewing candidates in an accept/reject loop, tuning the ranker, and persisting per-job memory — using natural language.

Who it's for: Recruiters, agencies, founders, and hiring teams who already use Pin.
What you connect: Claude Desktop, Claude Code, ChatGPT, Codex, and any MCP-compatible client.
How it authenticates: OAuth 2.1 via WorkOS — one click through your pin.com account. No API keys to rotate.
What it runs on: A hosted, stateless Streamable HTTP endpoint at mcp.pin.com/mcp. No local install.
What it exposes: 10 tools across 4 phases: kickoff, review, tuning, and per-job memory.
What it costs: Free on every paid Pin plan. No add-on fee, no metered call limit.
Data handling: SOC 2 Type 2. Every call is workspace-scoped and written to the audit trail.

The 10 tools

Everything the assistant can actually do.

Four small phases, ten small tools. The assistant sets up a search, reviews candidates with you, tunes on the fly, and remembers what it learned for next time. You still approve every accept, reject, and outreach.

01 · Set up the search

list_jobs

List active candidate searches in your workspace. Used to resolve a role the recruiter mentions by name before any other call.

“Pull up my open searches — I think we have one for a staff PM.”

scrape_job_description

Fetch a job posting from Greenhouse, Lever, Ashby, or any public job board URL. Returns the title, location, and full description.

“Grab the JD from this Greenhouse link and start a search.”

create_job

Kick off a new candidate search from a title, description, and company. Sourcing runs in the background; call get_candidates right after.

“Start a search: Senior iOS engineer at Acme, Swift + SwiftUI, NYC, remote OK.”

02 · Review candidates

get_candidates

Return up to 10 top-ranked, unreviewed candidates for a job. Each batch ships with fit hints, pattern analysis, and anything the assistant has learned so far.

“Show me the next batch — 10 at a time is fine.”

accept_candidate

Advance a candidate. Kicks off outreach from your Pin-configured sequence. Always confirmed — the assistant never sends silently.

“Accept Maya — she's exactly the shape.”

reject_candidate

Decline a candidate with a standard or free-text reason. Sourcing quietly backfills a replacement so the queue never runs dry.

“Pass on Robert — wrong stack. Replace him.”

03 · Tune the search

modify_search

Adjust the search with plain-language instructions — location, seniority, skills, comp, anything. Applies before the next batch.

“Only candidates in the NYC metro, drop anyone below staff level.”

recalibrate_search

Feed accumulated accept/reject signal back into the ranker, then pull a fresh, better-matching batch. Auto-suggested when patterns emerge.

“You've rejected 5 junior people in a row — recalibrate.”

04 · Persistent memory

read_job_memory

Load what the assistant learned about your taste for this role in prior sessions — accepted signals, decline reasons, current confidence.

“What do you already know about how I evaluate for this one?”

update_job_memory

Persist the running summary — patterns spotted, predicted criteria, confidence level — so the next session doesn't re-learn from scratch.

“Remember: never below 5 yrs of production Swift; de-rank bootcamp-only.”

Install · 60 seconds

Connect Pin to your AI of choice.

One OAuth click and the Pin tools appear in your assistant. Streamable HTTP, remote hosted: no local server, no API keys, no config drift.

1 Open Claude Desktop → Settings → Developer → Edit Config.
2 Paste this block into claude_desktop_config.json.
3 Restart Claude. Approve the OAuth window when it opens.
4 Try: "Start a search for a senior iOS engineer."
Full docs at docs.pin.com/pin-mcp

~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "pin": {
      "type": "http",
      "url": "https://mcp.pin.com/mcp"
    }
  }
}