Model Context Protocol (MCP): Definition for Recruiters

The Model Context Protocol (MCP) is an open standard that lets AI assistants, such as Claude and ChatGPT, connect to external applications and data through a common interface, so the assistant can read information and take actions in those apps with the user's permission.

Anthropic first published MCP in November 2024, and it has since been adopted by other AI providers and many software vendors. For recruiters, it means an assistant you already use can work with your recruitment tools directly, instead of you copying and pasting between windows.

How MCP works, in plain terms

• An MCP server is run by the application you want to connect, for example a recruitment tool. It publishes a list of tools: specific actions the assistant may take, such as "list submissions" or "get submission status".
• An MCP client is built into the AI assistant. When you add a connector, the assistant discovers the tools and can call them while you chat.
• Authorisation is usually handled with OAuth: you sign in to the application, see which permissions the assistant is asking for, and approve or deny. You can revoke access later.

A remote MCP server is reached over the internet at a URL, so you add it to your assistant much like installing an app.

What MCP could do for recruiters

• Ask "which submissions are waiting for review?" and get an answer from your formatting tool.
• Check the status of an export without opening another tab.
• Pull a candidate's evidenced skills into a draft message, where your data policy allows it.
• Trigger an export of a CV that a recruiter has already approved.

Safety questions to ask any MCP connector

MCP makes it easy to give an assistant real power, so check:

1. Which actions are exposed? Read-only tools are lower risk than tools that change or send data.
2. Can the assistant approve or send anything? For client submissions, a human should stay in control of approval.
3. What candidate data reaches the AI provider? Anything a tool returns is processed by the assistant's provider.
4. How is access scoped and revoked? Look for per-permission scopes, short-lived tokens and an easy way to disconnect.
5. Are documents treated as data? Text inside a CV must never be able to instruct the assistant to do something.

How CVPitch uses MCP

CVPitch runs a remote MCP server at https://cvpitch.app/mcp, with OAuth sign-in for Claude and ChatGPT. Assistants can search and list submissions, check status, read evidence only where your workspace's sharing policy allows it, create draft previews and upload links, export submissions a recruiter has already approved, and suggest changes for human review. They can never approve a submission. Several other CV formatting vendors also advertise MCP support. Read our guide to connecting recruitment software to Claude and ChatGPT, and see integrations and developers.

Related terms

• Recruitment CRM: a common system to connect to an assistant.
• Applicant tracking system: another system assistants can work with through MCP.
• CV parsing: the extraction step whose output assistants may read.

Want to see it on your own CVs? Start a free trial with 10 exports and no card, compare plans on the pricing page, or browse all features and integrations.