Cookies on this site

Strictly necessary cookies keep the site working. Our analytics and advertising tags — Microsoft Clarity and Google Tag Manager — stay switched off, and write no cookie, until you accept them. Privacy Policy

WhatsApp
Book free demo Start free
Open platform · MCP and API

Point your own AI at your hiring system.

Pitch N Hire speaks MCP. Connect Claude, ChatGPT, Gemini or an agent your team wrote, and it can read requisitions, rank candidates, schedule interviews and post jobs — under exactly the permissions the human whose key it uses already has. Nothing about your AI strategy has to run through us.

The same interfaces reach the whole record: the applicant tracking system and the HRMS, not a read-only slice of either.

Free sandbox, no card. The same endpoints in sandbox and in production.

MCP Native server, not a community wrapper REST Documented API over the same objects Hooks Signed webhook events, replayable 0 Of your data used to train shared models
MCP Claude Desktop — MCP configuration
{
  "mcpServers": {
    "pitchnhire": {
      "command": "npx",
      "args": ["-y", "@pitchnhire/mcp-server"],
      "env": {
        "PNH_API_KEY": "pnh_live_…",
        "PNH_WORKSPACE": "acme-talent"
      }
    }
  }
}

# then, in the assistant:
> "Rank the Bengaluru backend pipeline against
>  the approved scorecard and tell me which two
>  candidates the panel disagreed on."

called pnh.requisitions.list
called pnh.candidates.rank   (812 -> 34)
called pnh.scorecards.get
-> 34 clear the bar. Panel dissent on 2:
   REQ-4471/c_88213 — design depth
   REQ-4471/c_90114 — systems depth
   Reasoning and evidence attached per score.

Add the server once. Your assistant then has every tool your permissions allow.

REST REST API — rank a pipeline
curl -X POST https://api.pitchnhire.com/v1/requisitions/REQ-4471/rank \
  -H "Authorization: Bearer $PNH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
        "scorecard_id": "sc_backend_sr",
        "explain": true,
        "limit": 25
      }'

# 200 OK
{
  "requisition": "REQ-4471",
  "ranked": 25,
  "pool": 812,
  "results": [
    {
      "candidate_id": "c_88213",
      "score": 91,
      "evidence": [
        { "criterion": "distributed systems",
          "met": true,
          "source": "resume:p2:l14" },
        { "criterion": "team leadership",
          "met": "partial",
          "source": "interview:i_5521:14:22" }
      ],
      "decision_required_by": "human"
    }
  ]
}

Same objects, same permissions, cursor pagination and idempotency keys throughout.

Webhooks Webhooks — react in your own stack
// Verify the signature, then do whatever you like.
import { verify } from "@pitchnhire/webhooks";

export async function POST(req) {
  const raw = await req.text();
  const event = verify(raw, req.headers.get("pnh-signature"),
                       process.env.PNH_WEBHOOK_SECRET);

  switch (event.type) {
    case "candidate.ranked":
      await ourAgent.review(event.data);       // your model
      break;
    case "interview.completed":
      await slack.post(summarise(event.data)); // your tools
      break;
    case "offer.accepted":
      await payroll.createEmployee(event.data);
      break;
  }
  return new Response("ok");
}

Signed with HMAC-SHA256, retried with backoff, and replayable from the dashboard.

01 Four ways to use it

Use our AI, bring your own, or run us headless.

Our agents, on by default

Built-in agents source, screen, schedule, draft and grade. Nothing to configure, and evidence is attached to every score.

Most teams start here

Your assistant, via MCP

Claude, ChatGPT or Gemini operates your workspace directly — reading pipelines, ranking, scheduling and drafting — inside your permissions.

Teams with an AI habit already

Your model, via API

Call your own model on our data, or ours on yours. Bring a fine-tuned screening model and we will run the record around it.

Engineering-led talent teams

Headless

Keep us as the system of record and build your own interface or orchestration on top. MCP, REST and webhooks cover everything the interface does.

Platform and GCC teams

02 What an agent can do

Every tool a recruiter has, exposed as a tool an agent can call.

The MCP server covers the whole record. An agent cannot reach anything the key's owner cannot, and every call is written to the same audit log as a human action.

01 requisitions.list / get / create Open roles, approval state, budget band, scorecard link Read + write
02 candidates.search / rank Search your database, or rank a pipeline with evidence per score Read
03 candidates.advance / reject Queues the decision — a named human confirms it Human-gated
04 interviews.schedule / reschedule Real availability, panel briefs, candidate self-booking links Read + write
05 interviews.transcript / scorecard Transcripts, scores and evidence spans per competency Read
06 messages.draft / send Outreach and candidate replies; sends respect your consent settings Read + write
07 jobs.publish Post to connected boards and your own career site Read + write
08 vendors.scorecard Fill rate, submission quality, time to submit, effective margin Read
09 employees.create Offer accepted, through to employee record and payroll handover Human-gated
10 audit.export Complete decision trail for a requisition, regulator-ready Read
03 Governance

An agent with your permissions, not a back door.

Keys belong to people

An agent key is issued to a user and inherits their role, workspace and field-level permissions. Revoke the person, revoke the agent.

One audit log

Agent actions land in the same immutable log as human ones, tagged with the model and the key that made them.

Human-gated actions

Advance, reject, offer and employee creation always need a named person to confirm — by design, not by policy.

No shared training

Your data never trains a model used by anyone else, and model providers operate under zero-retention terms.

Predictable limits

Every key has a published per-minute and per-day limit with a burst allowance, and a 429 response tells you exactly when to retry.

Free sandbox

A full-fidelity sandbox workspace with seeded data, so you can build against it before you buy anything.

Data residency, retention and the rest of the controls are covered on the security page.

Works with Claude DesktopClaude APIChatGPT connectorsGemini function callingCursorContinueLangGraphCrewAIn8nZapier
04 What teams build with it

Four patterns the open platform is built for.

Morning pipeline briefing in Slack

An agent reads every open requisition overnight, flags the ones that stalled and drafts the nudge to each hiring manager.

MCP · webhooks · Slack

Your own screening model, our record

A platform team runs a fine-tuned model for niche engineering roles and writes its scores back onto the candidate record with evidence.

REST · candidates.rank

Career site as a custom app

Headless: the site is yours, applications and status live in Pitch N Hire, and candidates never see two systems.

REST · headless

Compliance pack for a regulator request

An agent assembles the full decision trail for any requisition — scores, evidence, who decided, when — as one export.

MCP · audit.export

05 Questions

Questions about the open platform.

What is MCP, in one sentence?

The Model Context Protocol is an open standard that lets an AI assistant call tools in another system. Because Pitch N Hire ships an MCP server, an assistant like Claude or ChatGPT can work inside your hiring data directly, rather than you copying text between windows.

Can an agent do something a recruiter could not?

No. An agent authenticates with a key issued to a specific user and inherits that user's role and workspace permissions exactly. If the person cannot see salary bands or delete a candidate, neither can their agent.

Is an agent allowed to reject candidates?

Not on its own. Advance and reject are the two actions that always require a named human, whether they come from the interface, the API or an agent. An agent can recommend and queue a decision; a person confirms it, and the audit log records who.

Do you train on our data if we use our own model?

We never train shared models on your data, whichever route you use. When you bring your own model, prompts and responses go straight between your provider and your workspace; we store the request metadata for the audit log and nothing else.

What are the rate limits, and are they enough for an agent?

Every key has a published per-minute and per-day limit with a burst allowance, and a 429 response tells you exactly when to retry. Agentic workloads are mostly reads, so the default ceiling is sized for continuous operation by a large team, and enterprise plans lift it on request. Ask us for the current figures for your plan.

Can we run Pitch N Hire headless?

Yes. Everything that matters is reachable through MCP, REST and webhooks, so you can keep us as the system of record while your own interface or agents do the work. Running your own career site and internal tooling on top of it is a supported pattern.

Which assistants does it work with?

Any client that speaks MCP or supports function calling — Claude, ChatGPT connectors, Gemini, editor clients such as Cursor and Continue, and orchestration frameworks such as LangGraph and CrewAI. Because the interface is an open protocol rather than a per-vendor integration, a new client that speaks MCP works without anything changing here.

The data your assistant reads comes from the rest of the platform: the ATS for pipelines and candidates, interview intelligence for scores and evidence, vendor management for agency submissions, and HRMS once someone is hired. See also integrations and how we handle access and security.

Give your AI a hiring system it can actually operate.

Sandbox keys are free. Or bring your CTO to a call and we will connect your own agent to a live workspace while you watch.

Start free Book demo