Workable MCP server icon

Workable

by Workable

Productivity45 tools

Run an applicant tracking system from your AI assistant. 45 listed tools, 94 documented — and the connector can reject a real job applicant, approve an offer and submit a performance review.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Workable via MCP

https://mcp.workable.com/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Workable Tools & Capabilities (45)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • An agent with w_candidates can reject applicants. Workable's scope model does not separate disqualification from routine candidate writes. This is the single most consequential fact on this page: scope the grant, or gate the tool in your client, rather than assuming a stage-move permission is bounded.
  • Candidate records are sensitive personal data. The tools return names, contact details, resumes, cover letters, application answers and attached files — get_candidate_files returns temporary pre-signed links, and get_candidate exposes a resume_url and per-answer file_url values. Workable's own guide warns that the employee scope grants access to confidential employee data. Treat any agent transcript touching these tools as holding personal data.
  • The published tool list is a floor. 49 documented tools sit outside Anthropic's 45, and 24 of them mutate. Scope a deployment to what the OAuth grant and the signed-in Workable user can reach, not to the directory's list.
  • Some writes are documented as final. submit_review fails on a second submit, mark_review_task_ready is described as a point of no return once a cycle is active, and sign_review is final. Workable's own docs tell an agent to confirm with the user before each.
  • Full-replace semantics bite. update_candidate_tags replaces the entire tag set rather than appending, and update_review_form replaces a whole questionnaire rather than merging. An agent patching conversationally will silently drop what it did not mention.
  • Rate limits are tight. 50 requests per 10 seconds on an OAuth token, 10 per 10 seconds on an account token, then HTTP 429. An agent walking a long candidate list will meet this.
  • Plan and token-type gates apply. Workable's REST reference records requisition endpoints as requiring the Hiring Plan, and offer endpoints as reachable only with user tokens.
  • No streaming. Workable documents the transport as stateless Streamable HTTP in JSON mode, with no SSE.
  • Workable does not support CORS, and states that resource IDs exposed through the API differ from those in the application for non-disclosure reasons — so IDs an agent reports will not match what a colleague sees in the Workable UI.
  • We could not read tool schemas or safety annotations. The endpoint requires OAuth, so no parameter types and none of MCP's readOnlyHint / destructiveHint annotations were observed. The read/write split here is counted from tool names and Workable's own scope mapping, not from annotations.

Frequently asked questions

Can the Workable MCP server reject a job applicant?

Yes. Workable documents a disqualify_candidate tool that disqualifies a candidate with a reason of rejected or withdrew, and a companion get_disqualification_reasons that the docs describe as required before disqualifying. Anthropic's directory lists both. Disqualification is reversible through a documented revert_disqualification tool, but the applicant-facing consequence of a rejection is not something a revert undoes.

Does the Workable MCP server send email to candidates?

No. Across all 94 tools Workable documents, none sends a message to a candidate. The only tool that sends anything is share_review, which shares a completed performance review with an employee and notifies them, and the only tool that emails anyone is invite_member, which invites a colleague to the Workable account. Candidate correspondence stays in Workable's own interface.

Can an agent post or unpublish a job through Workable MCP?

No. Every one of the nine job tools Workable documents is a read: list jobs, search jobs, fetch a job, its stages, its members, its recruiters, its application form, its custom attributes and its activity log. Workable's REST reference publishes no create, update, publish or unpublish endpoint for jobs either, so the capability does not exist to expose.

How many tools does the Workable MCP server actually have?

Anthropic's directory lists 45 tool names, but Workable's own MCP reference documents 94 across seventeen sections. All 45 listed names appear in that documentation, so the directory is a strict subset rather than a contradiction. The 49 undocumented-by-Anthropic extras include 24 mutating tools, among them relocate_candidate, update_candidate, submit_review and delete_member.

Can you connect Workable MCP with read-only access?

Yes, in principle. Workable's RFC 9728 descriptor advertises 21 scopes split by resource family, and eleven of them are read scopes prefixed r_ that carry no write counterpart in a single grant. Whether your MCP client lets you decline individual write scopes at the consent screen is a client question rather than a Workable one.

Which scope controls candidate rejection in Workable?

The same one that controls every other candidate write: w_candidates. Workable's REST reference maps disqualify, revert, copy, relocate, move, rate and update all to that single scope. There is no narrower grant that permits sourcing or stage moves while withholding rejection, so an agent given w_candidates for any reason can disqualify applicants.

Is there a rate limit on the Workable MCP server?

Yes. Workable documents 50 requests per 10-second interval for OAuth 2.0 tokens, with excess requests returning HTTP 429. Account tokens get 10 requests per 10 seconds instead. Every response carries X-Rate-Limit-Limit, X-Rate-Limit-Remaining and X-Rate-Limit-Reset headers, and the limit is scoped per client rather than per endpoint.

Does the Workable MCP server need an account parameter?

Yes, on every tool but one. Workable documents that each tool except get_accounts requires an account parameter, so an agent must call get_accounts first and pass the returned subdomain on every subsequent call. That tool is documented by Workable but absent from Anthropic's 45-name listing, which makes the required first step invisible in the directory.

Sources

Use in Agentman

Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Server Info

Category
Productivity
Developer
Workable
Tools
45
Domain
mcp.workable.com

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.