ENTRA MCP — the first agent-ready job platform

Let your AI find your job. This MCP server connects Claude, ChatGPT, Cursor, Windsurf or any MCP client — locally via npx or hosted at https://mcp.entracareers.com/mcp — to ENTRA — thousands of verified AI & tech jobs aggregated straight from company hiring systems (OpenAI, Anthropic, SpaceX, Stripe + 300 more). Zero ghost jobs.

Agents search, match, rank and prepare. Humans decide and apply — no spam, no auto-submission.

Install

Claude Code

claude mcp add entra -- npx -y entra-mcp

Claude Desktop — one-click extension (.mcpb) Download entra-mcp.mcpb and open it with Claude Desktop (Settings → Extensions → Install from file). No Node install needed. Employers can paste their API key in the extension settings.

Claude Desktop (claude_desktop_config.json), Cursor (.cursor/mcp.json), Windsurf (~/.codeium/windsurf/mcp_config.json)

{
  "mcpServers": {
    "entra": { "command": "npx", "args": ["-y", "entra-mcp"] }
  }
}

From source:

npm install && npm run build
claude mcp add entra -- node "/path/to/entra-mcp/dist/index.js"

Requires Node 18+. No API key needed for candidate tools — they are read-only against the public ENTRA API. Employers add one env var, see Employer mode.

Then just ask: "Find me remote ML engineer roles paying $200K+, tell me which companies are hiring most, and prepare an application for the best fit."

Hosted endpoint (no install)

The same server runs at https://mcp.entracareers.com/mcp (MCP Streamable HTTP). Nothing to install, no account for candidate tools.

Employer mode over HTTP: send your key on every request as Authorization: Bearer entra_live_… and the 8 employer tools appear for that request (Cursor: add "headers": { "Authorization": "Bearer entra_live_…" } next to url). ChatGPT and claude.ai connectors cannot send custom headers today, so there you get candidate mode — employers use the local npx setup with ENTRA_API_KEY (Employer mode).

How the hosted server behaves:

• Stateless: every request stands alone (no sessions, no Mcp-Session-Id), so it scales and redeploys without breaking clients. GET /mcp and DELETE /mcp answer 405 as the spec allows.
• The Authorization header is used for that one request and is neither stored nor logged. No key is read from the server's environment.
• Rate limit 120 requests/minute per client IP on /mcp; request bodies up to 1 MB; CORS open for browser-based agents. Health: GET /healthz.
• Same honesty rules as below — the hosted server is just another transport for the identical tools.

Tools

Every job in every response carries verified: true, a salary band when the company lists one, and an apply URL of the form https://entracareers.com/{country}/vacancies/{id}?utm_source=agent&utm_medium=mcp — we welcome agent traffic.

Sample: match_jobs

{
  "source": "ENTRA — verified roles aggregated from company ATSs. Finds and ranks; you apply.",
  "profile": { "skills_used": ["machine learning", "python", "rag", "llm", "pytorch", "aws"], "target_role": "machine learning engineer", "seniority": "6+ yrs (7+ years mentioned)" },
  "queries_run": ["machine learning engineer", "python", "rag", "llm"],
  "candidates_considered": 142,
  "matches": [
    {
      "title": "Machine Learning Engineer, Inference",
      "company": "OpenAI",
      "location": "San Francisco, United States",
      "work_location": "remote",
      "salary": "$295K–555K/yr",
      "url": "https://entracareers.com/us/vacancies/…?utm_source=agent&utm_medium=mcp",
      "fit_score": 88,
      "interview_chance_hint": "high",
      "why": [
        "Skills: 5/6 matched (machine learning, python, llm, pytorch, aws)",
        "Title match: \"Machine Learning Engineer, Inference\" fits \"machine learning engineer\"",
        "Seniority: job asks 6+ yrs, you read as 6+ yrs — match",
        "Remote role (as requested)",
        "Salary band listed: $295K–555K/yr"
      ],
      "skill_gaps": ["rag"]
    }
  ]
}

Honesty note

• Matching is deterministic — a keyword dictionary plus rules, no LLM in the loop. fit_score is explainable, not a prediction; interview_chance_hint is derived from fit_score only (never a made-up percentage).
• salary_stats uses listed bands only — roles without a published band are excluded, and nothing is estimated.
• search_jobs rank_by: "salary" returns salaried listings only; posted_within_days is applied client-side over the most recent 500 matches.
• Nothing is submitted on your behalf. prepare_application returns a link and a checklist; the human applies.

Employer mode (API key)

Same package, one env var: with ENTRA_API_KEY set, the 8 candidate tools stay and 8 employer tools are added — post jobs, manage listings, read applications, search and match candidates for your company.

1. Create a key at entracareers.com/employer/profile?tab=api-keys (pick the scopes you need; the key is shown once).
2. Connect:

Claude Code

export ENTRA_API_KEY="entra_live_…"
claude mcp add entra -e ENTRA_API_KEY=$ENTRA_API_KEY -- npx -y entra-mcp --employer

Claude Desktop / Cursor / Windsurf

{
  "mcpServers": {
    "entra": {
      "command": "npx",
      "args": ["-y", "entra-mcp", "--employer"],
      "env": { "ENTRA_API_KEY": "entra_live_…" }
    }
  }
}

Then ask: "Post a Senior Sales Manager job in Dubai — hybrid, 15–22K AED/month, 45 days — then show me the 5 best candidates for it and who has already applied."

How it behaves:

• The key is read from the environment only (never from arguments). Without a key the employer tools are not registered; --employer alone just prints a hint on stderr.
• Nothing is called at startup. The first employer call resolves your company and scopes via GET /employer/api-keys/me and caches them.
• Only employer endpoints receive the bearer header; public lookups (countries, cities, specializations, job details) are made anonymously.
• Rate limit is 120 requests/minute per key; a 429 is retried once after retry-after.

Employer tools