Use CVPosh from ChatGPT, Claude, Cursor and Gemini

CVPosh has an MCP server built for AI agents. Your agent reads the CVs you saved in CVPosh, scores them for applicant tracking system (ATS) compatibility and compares them with a job posting. You tag the work, approve the connection and review what the agent reports; every tool is read-only, so you make each change yourself in the editor. Markdown version of this page: /mcp.md.

Agent quickstart

Discovery

• Short description and links: https://cvposh.com/llms.txt
• Full reference for agents (this quickstart included): https://cvposh.com/llms-full.txt
• MCP server card (URL, auth, tools, limits): https://cvposh.com/.well-known/mcp/server-card.json
• Official MCP Registry name: com.cvposh/cvposh
• OAuth metadata: https://cvposh.com/.well-known/oauth-protected-resource
• Markdown pages: https://cvposh.com/mcp.md and https://cvposh.com/pricing.md

Authentication

• There is no agent self-signup. A human owns a CVPosh account and either approves an OAuth connection or creates an API key.
• OAuth 2.1 with PKCE and dynamic client registration. Redirect URIs are limited to ChatGPT's and Claude's published callbacks and http://localhost or http://127.0.0.1 loopback addresses. The human signs in and clicks Allow on the consent page.
• API key: the human creates one under Settings, then API keys, and gives it to the agent to send as "Authorization: Bearer <key>". This is the route for clients whose callback is not accepted, such as Cursor.

Minimal call sequence

1. 1. initialize, then tools/list: POST one JSON-RPC 2.0 message per request to the server URL. The server is stateless: no session id.
2. 2. list_cvs: Returns each saved CV's id, title, template and last-updated time, with no CV content.

   {"name":"list_cvs","arguments":{}}

3. 3. ats_check: Scores one saved CV (cvId) or text you pass (resumeText), never both: 0 to 100 with issues, suggestions and stats.

   {"name":"ats_check","arguments":{"cvId":"<id from list_cvs>"}}

4. 4. analyze_keywords: Compares a CV or text with a job posting (at least 50 characters): matched keywords, missing keywords and a match score.

   {"name":"analyze_keywords","arguments":{"cvId":"<id from list_cvs>","jobDescription":"Senior data analyst. SQL, dashboards, stakeholder reporting, A/B testing, Python..."}}

5. 5. get_cv (only when the content is needed): Returns one CV's sections as structured data and plain text, without the profile photo. Long output is cut and marked truncated.

   {"name":"get_cv","arguments":{"cvId":"<id from list_cvs>"}}

6. 6. Hand off to the human: Report what you found and what you suggest. You cannot edit the CV: the human changes it in the editor at https://cvposh.com/builder and decides which suggestions to apply.

Limits, retries and idempotency

• 120 requests a minute and 100 tool calls a day per connection (the day resets at 00:00 UTC).
• 60 unauthenticated requests a minute per IP address; 1200 requests a minute per IP address in total.
• Resume and job text up to 50,000 characters; request bodies up to 256 KB.
• There is no paid-plan gate on the tools, so no "plan required" error exists. Idempotency: every tool is read-only and declared idempotent, so repeating a call is safe.

Errors the agent should handle

Human oversight

What you, the human, can see, approve, limit and revoke today:

Not available today: No per-call activity log: Settings shows a last-used date, not which tool was called with what. No notification when a connection is created or used. No agent write tools, so there is nothing to put in an approval queue; an agent cannot edit or tailor a CV itself. No MCP tools for the employer (hiring) workspace. Your assistant may also ask you to approve each tool call; that is a setting of the client, not of CVPosh.

Tools

The ATS score is a fixed checklist, not a simulation of any employer's system and not a prediction of interviews. See how the ATS checker scores. Server URL: https://cvposh.com/api/mcp. Registry name: com.cvposh/cvposh in the Official MCP Registry.

Connect in 60 seconds

claude mcp add --transport http cvposh https://cvposh.com/api/mcp

/plugin marketplace add airdropfi/cvposh-claude-plugin
/plugin install cvposh@cvposh-plugins

{
  "mcpServers": {
    "cvposh": {
      "url": "https://cvposh.com/api/mcp",
      "headers": { "Authorization": "Bearer ${env:CVPOSH_API_KEY}" }
    }
  }
}

gemini extensions install https://github.com/airdropfi/cvposh-gemini-extension
/mcp auth cvposh

{
  "inputs": [
    { "type": "promptString", "id": "cvposh_key", "description": "CVPosh API key", "password": true }
  ],
  "servers": {
    "cvposh": {
      "type": "http",
      "url": "https://cvposh.com/api/mcp",
      "headers": { "Authorization": "Bearer ${input:cvposh_key}" }
    }
  }
}

{
  "mcpServers": {
    "cvposh": {
      "serverUrl": "https://cvposh.com/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_CVPOSH_API_KEY" }
    }
  }
}

Prompts to copy

Written for the human's part of the job: you assign the work and review the result. Each prompt works with the four read-only tools.

Sign-in and data handling

Claude, ChatGPT and local clients sign in with OAuth 2.1 (PKCE): you sign in on cvposh.com and click Allow, and the assistant never sees your password. Clients that cannot use OAuth, such as Cursor, send an API key from Settings › API keys as a Bearer header. Each connection is listed there; delete it to disconnect. What an assistant reads is sent to that assistant's provider and handled under its privacy policy. Full details are in the privacy policy.

Frequently asked questions

Prefer a browser? Use the free ATS checker or the CV builder. Prices are on the pricing page.