Ashby候选人评估MCP服务

ashby-mcp

MCP server for Ashby ATS — designed for candidate evaluation workflows.

This server exposes Ashby's recruiting data through the Model Context Protocol, letting AI agents review candidates, read application details, and write evaluation notes.

Prerequisites

You need an Ashby API key:

1. Go to your Ashby admin settings → Integrations → API Keys
2. Create a new API key with these permissions:
   • candidatesRead — read candidate profiles, applications, notes, feedback
   • jobsRead — read job listings and details
   • interviewsRead — read interview stages and plans
   • candidatesWrite — add notes, tags, move application stages, archive applications
   • hiringProcessMetadataRead — list archive reasons and email templates

Optional permissions for the raw API tool:

• apiKeysRead — inspect the current key with apiKey.info.
• sourcingRead — read sequence templates, email senders, and existing sequences.
• emailsRead — read recent candidate email messages.

A 403 with missing_endpoint_permission requires updating the key's permissions in Ashby Admin > Integrations > API Keys. The MCP reports the endpoint and required permission for these operations. Email and sequence endpoints may also require beta access for your organization; see the Ashby authentication reference and the endpoint's documentation.

Install

Claude Code (recommended)

Run in your terminal:

claude mcp add ashby -e ASHBY_API_KEY=your-api-key-here -- npx -y ashby-mcp@latest

This auto-updates whenever a new version is published.

Claude Desktop

Add this to your Claude Desktop MCP config (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "ashby": {
      "command": "npx",
      "args": ["-y", "ashby-mcp@latest"],
      "env": {
        "ASHBY_API_KEY": "your-api-key-here"
      }
    }
  }
}

This also auto-updates on each restart.

Alternative: You can download the .mcpb bundle and double-click to install — no terminal needed. Note that this method pins you to a specific version and won't auto-update. You'll need to re-download after each release.

Other MCP clients

Add the same JSON config above to your client's MCP server configuration.

Docker

Build the image, then add it to your MCP client config:

docker build -t ashby-mcp .

{
  "mcpServers": {
    "ashby": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "-e", "ASHBY_API_KEY", "ashby-mcp"],
      "env": {
        "ASHBY_API_KEY": "your-api-key-here"
      }
    }
  }
}

From source

git clone https://github.com/dewierwan/ashby-mcp.git && cd ashby-mcp
npm install
npm run build

Then point your MCP client at node dist/index.js with the ASHBY_API_KEY environment variable set.

Tools

All tools return dual-format responses: a human-readable summary followed by structured JSON.

Jobs

ashby_list_jobs

List open jobs with IDs, titles, department, location, and status.

ashby_get_job_details

Full job details including description and interview plan stages.

ashby_get_pipeline_summary

Pipeline overview with candidate counts per stage, per job.

Candidates

ashby_get_candidate

Comprehensive candidate profile with all applications resolved.

ashby_get_candidate_notes

All notes on a candidate.

ashby_search_candidates

Search candidates by name or email.

ashby_add_candidate_note

Add an evaluation note to a candidate. Visible to the hiring team.

ashby_add_candidate_tag

Tag a candidate (e.g. "Strong Hire", "Needs Review").

ashby_get_resume

Download and extract text from a candidate's resume (PDF, text).

Applications

ashby_list_candidates_for_job

List candidates/applications for a specific job with stage and status.

ashby_get_application_details

Full application with stage history, feedback, and criteria evaluations.

ashby_get_application_form_submission

Candidate's submitted application form responses (screening questions, cover letter, etc).

ashby_list_applications

List applications across all jobs with date, status, stage, and source filters.

ashby_archive_application

Archive an application with reason and optional rejection email.

ashby_bulk_archive

Archive multiple applications at once (max 25).

ashby_archive_by_filter

Find and archive every application matching a set of filters in one call. It pages through application.list itself, so you don't need to collect IDs first. Use it for bulk cleanups instead of ashby_bulk_archive.

Filters combine with AND. At least one filter is required, so the tool can never run unfiltered.

Hired and already-archived applications are never touched. Archives run five at a time, with extra backoff on 429 and 5xx responses. One failure never stops the run.

The response is compact: totals (matched, to archive, archived, skipped by reason, failed), a per-job breakdown, up to 10 sample names per job in a dry run, and every failure with its error. It does not list archived IDs.

Safe workflow: always run with dry_run: true first (the default). Check the totals, per-job counts and sample names, then repeat the same call with dry_run: false.

Dry run: preview every lead on jobs closed before 4 July 2026.

{
  "job_closed_before": "2026-07-04",
  "application_status": ["Lead"]
}

Real run: archive them, keeping two candidates and allowing up to 1,500 archives.

{
  "job_closed_before": "2026-07-04",
  "application_status": ["Lead"],
  "exclude_candidate_ids": ["<candidate-uuid>", "<candidate-uuid>"],
  "archive_reason_id": "<reason-uuid>",
  "dry_run": false,
  "max_applications": 1500
}

Interviews

ashby_list_interview_stages

List all interview stages across all interview plans. No parameters.

ashby_list_upcoming_interviews

List upcoming and pending interview schedules.

Workflow

ashby_move_application_stage

Move an application to a different interview stage.

ashby_get_feedback

Submitted feedback/scorecards for an application.

ashby_list_archive_reasons

List available archive/rejection reasons. No parameters.

ashby_list_email_templates

List email templates for rejection emails. No parameters.

Escape hatch

For anything the dedicated tools above don't cover (e.g. offer.list, user.list, department.list, interviewPlan.list, candidate.update), use the two generic tools below. They share the same auth, retries, and timeouts as every other tool.

ashby_call_api

Call any Ashby API endpoint directly. Returns the raw response envelope, including success, results, moreDataAvailable, and error details.

Marked destructiveHint: true so Claude surfaces the call before running it (a generic passthrough can hit write endpoints).

ashby_get_api_docs

Return a curated reference for the Ashby API: base URL, auth, response envelope, pagination, endpoint index grouped by resource, and common gotchas. No parameters.

Call this before ashby_call_api if you're not sure which endpoint or parameter names to use. Full official reference: https://developers.ashbyhq.com/reference/.

Example Prompts

Show me all open engineering jobs.

List the candidates for the Senior Backend Engineer role and summarize where each one is in the pipeline.

Pull up the full profile for candidate Jane Smith — I want to see her resume info, application history, and any existing notes.

Review the feedback submitted for application abc-123 and summarize the interviewer evaluations.

Add a note to candidate xyz-456: "Strong technical skills demonstrated in system design round. Recommend advancing to final interview."

Move application abc-123 to the "Final Interview" stage.

Read the resume for candidate Jane Smith and summarize her experience.

How many people applied this week?

What does our pipeline look like for the Community Lead role?

Show me all candidates currently in the Work test stage.

Archive John Smith's application for the Senior Engineer role with reason "Not enough experience" and send the standard rejection email.

Show me the available archive reasons and rejection email templates.

Bulk archive all candidates in Application Review for the closed Designer role.

Dry-run archiving every lead on jobs closed before July, then archive them.

Privacy

• This server proxies requests to the Ashby API using your own API key.
• No data is stored, logged, or sent anywhere other than Ashby's API endpoints (api.ashbyhq.com).
• Your API key is stored locally on your machine and is never transmitted to any third party.

Testing

Verify API connectivity

ASHBY_API_KEY=your-key npm run test-api

Test with MCP Inspector

ASHBY_API_KEY=your-key npx @modelcontextprotocol/inspector node dist/index.js

Run directly

ASHBY_API_KEY=your-key node dist/index.js

The server communicates over stdio — it will start silently and wait for MCP protocol messages.

Development

npm install
npm run build    # compile TypeScript to dist/
npm test         # run tests
npm start        # run the built server