.webp?dpl=dpl_CEoS6wcaD39NGG2gyZSjANr3BAcG)
Which ATS can your AI agent actually operate?
Yardstick ships a hosted MCP server, a public API and a yardstick command-line tool on every account, so the agent you already work in can read where hiring stands and prepare the next piece of work. The decisions a candidate can see stay with you.
What the agent reads, and what it prepares.
You are already working in an agent. The question is whether your hiring system can be reached from inside it, or whether it stays a browser tab you leave the conversation to go visit.
In Yardstick it can be reached, and the model is worth being precise about. You have a real discussion with your own agent, where you bring the business context that is not written down anywhere. Your agent runs Yardstick to turn that discussion into work: an interview plan for the role as you actually described it, the questions, the scorecard, the drafts. What comes back is an artifact in Yardstick that you look at and change before it counts.
What your agent can read
Where a role stands. Jobs, job postings, hiring pipelines, and the applications sitting in each stage.
Who a candidate is and what has happened to them. The person record, the application, the job it is attached to, and the interviews scheduled against it.
What the panel is going to ask. The published interview plan, which is the questions and the written anchors for what each score means, plus the interviewers assigned to each session.
Whether the evidence is in yet. Who still owes feedback on a role, and the hiring decision brief once the assigned scorecards are complete.
Where your outreach stands. Prospects, prospect activity, talent pools, source and referral links, candidate email templates and their versions, and any candidate email drafts waiting for review.
Who changed what. API events and the audit log.
Every tool carries a human-readable title and read-only or destructive hints, so your assistant's consent screen tells you what a tool does before you approve it.
What it prepares, and what you sign off
| Your agent prepares | You approve before |
|---|---|
| Job descriptions | Publishing a job |
| Interview plans, questions and scorecards | Advancing or rejecting a candidate |
| Summaries of candidate evidence | Making or communicating a hiring recommendation |
| Scheduling and email drafts; prospect pipelines, talent pools and outreach sequences | Sending an ad-hoc candidate email outside an approved sequence |
| Candidate messages inside a sequence you have already approved | Activating an outreach sequence that sends candidate email |
| Changing permissions, API tokens or billing |
The discussion your agent turns into a plan can carry more than the role description. If you want the plan aimed at a specific gap on the team, context-aware hiring is how your team's metrics reach that same conversation. And the recurring work that runs on a schedule rather than on your ask is handled by the built-in agents, whose output lands in the same review queue you approve from.
Connecting it.
1. A coding agent. Install the
yardstickCLI and authenticate. Claude Code, Codex or a similar agent runs it for you. The CLI is agent-agnostic.2. A general assistant over MCP. Add Yardstick's hosted remote MCP server as a connector at
https://api.yardstick.team/mcpand sign in with OAuth. ChatGPT and Claude both connect this way. Setup is in the MCP documentation, and the MCP server page lists what its 176 tools cover and the connect steps for each client.3. Your own code. The public REST API is the same surface the other two run on, so a coding agent and a chat assistant share one system of record.
Access is scoped to what you can already see and tokens are short-lived. Over the MCP server, every write-capable tool requires an explicit idempotency key, so a retry cannot apply the same write twice.
All three ship on every account, including the free one. Pricing is pay-as-you-go: you pay for active hiring rather than for seats or an annual contract.
Who else publishes an agent surface.
We are not the only ATS you can reach from an assistant, and a page that pretended otherwise would not survive one search. Here is the field as each vendor describes it, in their own words.
Pricing, availability and status checked 2026-08-24 on each vendor's own site. Listed alphabetically. The order carries no opinion, including ours.
| ATS | Agent surface the vendor publishes | Status the vendor states | Plans the vendor states | What the vendor says about cost |
|---|---|---|---|---|
| Ashby | MCP server for ChatGPT, Claude, Cursor and other MCP clients; user-level OAuth | Open beta | All plans (Foundations, Plus, Enterprise) | No additional cost stated |
| Greenhouse | MCP server for Claude, ChatGPT, Google Antigravity, Glean, Copilot Studio, Amazon Q and Grok AI; OAuth 2.0 with PKCE; read and write | Currently in open beta | Core, Plus and Pro tiers | “Contact your account team to learn about access and pricing” |
| Lever | No first-party announcement found on the vendor's own site when we checked on 2026-08-24 | |||
| Workable | MCP server added as a custom connector in your AI tool; OAuth2; scoped to your own access level and assigned jobs | Shipped 2026-05-13, expanded 2026-07-20 | All plans | “Included on all plans at no additional cost” |
| Yardstick | Hosted MCP server at https://api.yardstick.team/mcp, public REST API, and a yardstick CLI a coding agent runs; OAuth, short-lived tokens | Live since 2026-06-20 | Every account, including the free tier | No additional cost; pay-as-you-go on open roles |
Two notes on reading this table honestly. Lever's row is blank because we looked and did not find a first-party announcement, which is a fact about our check rather than about their product. And no row carries a tool count, because a count changes the week any vendor ships more of them, and it tells you nothing about whether the agent can do the one thing you need.
Once you have narrowed the field, what each of these costs is the other half of the shortlist.
If you are already on one of these.
If you run Greenhouse or Ashby today, both publish a connector of their own, on the terms in the table above. An assistant reading the pipeline you already have is worth knowing about before you consider moving anything, and nothing below is worth a migration you did not otherwise want.
Yardstick is worth a look when the shape of the problem is different. Two cases.
You want a coding agent to run hiring, not a chat window. The yardstick CLI is a command-line tool, so Claude Code or Codex operates your hiring system the same way it operates the rest of your stack, and the public API is underneath both it and the MCP server.
You want the interview structure in the same place as the agent surface. The plan, the questions, the scorecard with its written anchors and the decision brief are the record your agent works against, so what it prepares comes back structured rather than as prose you have to file somewhere.
And it is on every account, including the free one, without a sales conversation first.
Where we are not the answer: if you need an HRIS, or you are standardising hiring across a thousand people, the enterprise platforms in the table above are the better starting point. If you are moving from Greenhouse, the switch-from guide covers the import.
The MCP server, the API and the CLI are on every account.
Questions people ask.
Does Yardstick have an MCP server?
Yes. It is a hosted remote MCP endpoint at https://api.yardstick.team/mcp, OAuth-protected and live on every account.
Which applicant tracking systems can an AI assistant connect to?
Checked on each vendor's own site on 24 August 2026: Workable publishes an MCP server included on all plans, Greenhouse publishes one in open beta for Core, Plus and Pro with access and pricing from your account team, Ashby publishes one in open beta on all plans, and Yardstick ships an MCP server, a public API and a CLI on every account.
Can ChatGPT or Claude operate my ATS?
In Yardstick, yes, over MCP. Add Yardstick as a connector and sign in, and the assistant reads your hiring data and prepares work in it. The actions a candidate can see wait for your approval.
Do I need a coding agent to use it?
No. A coding agent such as Claude Code or Codex uses the yardstick CLI, and a general assistant uses the MCP server. Both run on the same public API.
Can the agent email candidates on its own?
Candidate email sends inside a sequence you have approved, or as a single message you have approved. Nothing else.
Is any of this a paid or Enterprise feature?
No. The MCP server, the public API and the CLI are on every account, including the free one.
Put your hiring system where your agent already is.
A hosted MCP server, a public API and a yardstick CLI on every account, with the interview plans, scorecards and candidate evidence your agent works against. Your first three Jobs are free.