Connect Claude and ChatGPT to your recruitment software with MCP

You can connect Claude to recruitment software, and ChatGPT too, as long as the software offers an MCP server. MCP, the Model Context Protocol, is an open standard that lets an AI assistant use tools inside another app, with your permission and under your own login. For CVPitch, you add the server address https://cvpitch.app/mcp as a custom connector, sign in to CVPitch, approve the permissions, and then ask questions like "which submissions are waiting for review?". Two things matter more than the setup. Anything the connector returns to the assistant goes to that AI provider. And CVPitch's connector can never approve a submission. This guide covers both, plus the step-by-step setup.

What MCP is, in plain English

MCP was introduced by Anthropic in late 2024 as an open protocol, and it's now supported by many AI assistants, including Claude and ChatGPT. It works like this:

• An app publishes an MCP server, a list of tools (actions) with descriptions of what each one does.
• You add that server to your assistant as a connector. For remote servers like CVPitch's, you sign in to the app through OAuth, the same "sign in and allow" pattern you've used to connect other apps. The assistant never sees your password.
• When you ask a question, the assistant decides which tools to call. The app checks your permissions on every call and returns only what those permissions allow.

MCP isn't a back door into a database. The app's server decides what each tool can do, and that's where the safety comes from, not from the assistant behaving well.

Why recruiters are connecting assistants to their tools

The useful jobs are mostly about saving clicks and context-switching:

• Checking what's waiting for review, what's been exported and how much of this month's allowance is used.
• Pulling an approved document while drafting a client email in the assistant.
• Staging a suggested correction you've spotted, to review properly in the app.
• Combining status from several tools in one conversation.

The jobs to avoid are the ones where an assistant makes or implies a decision about a person.

The risks with candidate data, and how to manage them

• Data leaves your system. Whatever a tool returns becomes part of the conversation, processed by the assistant's provider under your account's terms and data settings. Treat sharing candidate content with an AI assistant as a separate, deliberate decision.
• CVs are untrusted input. A CV can contain text written to manipulate an AI ("ignore previous instructions…"). The app's permissions must hold whatever the assistant is told. In CVPitch, candidate documents and tool results never grant tool permissions.
• Over-broad access. Approve only the permissions you need, and revoke connections when someone leaves.
• Decisions about people. Keep humans approving submissions and making hiring-related decisions.
• Data protection duties. If candidate data reaches an AI provider, it needs to be covered in your records and privacy notice. Our general guide to GDPR and CV data covers processors. It isn't legal advice.

What CVPitch's MCP connector can do

Each capability needs a permission you approve when you connect. In plain English:

The exact tool list appears in your assistant when you connect, and we may add tools over time. Each one is bound by the same permissions.

What it can't do

• It can't approve a submission or change an approval, ever. Approval happens in CVPitch, by a person, on an exact revision.
• It can't change templates, team members, billing or retention settings.
• It can't read candidate content unless a workspace owner has enabled it.
• It can't send anything to clients.
• It can't exceed the permissions you approved. It acts as you, within your workspace, and you can revoke it at any time.

Step by step: connect Claude to recruitment software like CVPitch

Menu names change from time to time, so treat these as the general route and check Claude's help centre if a label differs.

1. Check your plan. You need a CVPitch account (the free trial works) and a Claude plan that supports custom connectors. On Team and Enterprise plans, an owner may need to add or allow the connector for the organisation first.
2. Open connector settings. In Claude, go to Settings → Connectors and choose to add a custom connector.
3. Enter the details. Name it "CVPitch" and use the remote MCP server URL https://cvpitch.app/mcp.
4. Connect and sign in. Select Connect. You'll be taken to CVPitch to sign in. The consent page shows your workspace and the permissions requested, and it states that the connection can't approve submissions. Choose Allow.
5. Use it in a chat. Back in Claude, turn the CVPitch connector on for your conversation from the tools menu, then ask a question.

Step by step: add CVPitch to ChatGPT

ChatGPT's support for custom MCP connectors depends on your plan and, in business workspaces, on admin settings. The general route:

1. Check availability. Make sure your ChatGPT plan supports custom connectors. In a business workspace, ask your admin whether custom connectors are allowed.
2. Enable custom connectors. In ChatGPT's settings, find the apps and connectors section. Where required, turn on developer mode under the advanced settings.
3. Create the connector. Add a new connector named "CVPitch" with the MCP server URL https://cvpitch.app/mcp, and choose OAuth as the authentication method.
4. Sign in and allow. Complete the CVPitch sign-in and review the requested permissions.
5. Select it in a conversation. Pick the CVPitch connector from the tools available in your chat.

Prompts to try

• "Which CVPitch submissions are waiting for review?"
• "How many exports have we used this month?"
• "Which client templates do we have, and what version is each on?"
• "Is the export for submission [ID] ready? Get me the download link."
• "Suggest a correction to the education dates on submission [ID]. I'll review it in CVPitch."

There's no approval tool, so the assistant can't approve anything, however the request is worded. That's by design.

Good practice for agency owners

• Start with metadata only. Status, usage and templates cover most of the everyday value.
• Enable candidate sharing deliberately, only if your data protection assessment supports it and your assistant plan's data controls suit candidate data.
• Review connections regularly, and revoke them when people leave.
• Treat assistant output as a draft. A summary or email from an assistant needs the same checking as anything else that reaches a client.
• Update your privacy notice if candidate data is shared with an AI provider.

Our security page explains how CVPitch handles access and storage.

Other ways to connect CVPitch

MCP is one of several routes. CVPitch has native connectors for ATSs and CRMs such as Greenhouse, Lever, Workable and Recruit CRM, a private email address you can forward CVs to, and a REST API with API keys and signed webhooks for status events. It can send notifications to Slack, Microsoft Teams, Discord and Google Chat, which carry statuses and links, never CV content. It connects to Zapier, Make and n8n through the API and webhooks, and it imports interview transcripts as .vtt files from Zoom, Microsoft Teams or Google Meet as labelled recruiter notes. See the integrations page and the developer documentation. If you're deciding where AI belongs in your workflow, our explainer on CV parsers and CV formatters is a useful starting point.

Frequently asked questions

What is an MCP server?

An MCP server is the part of an app that offers tools to AI assistants through the Model Context Protocol. The assistant can call those tools on your behalf, within the permissions you approved when you connected.

Can Claude or ChatGPT approve a CV in CVPitch?

No. The connector can export a submission only when a person has already approved that exact revision in CVPitch. It can't create, change or bypass an approval.

Is it safe to connect an AI assistant to candidate data?

It can be done responsibly, but it's a real data decision. Anything returned to the assistant is processed by its provider. Start with metadata-only access, enable candidate sharing only after a proper assessment, and keep humans approving submissions.

Do I need to be technical to set this up?

No. Adding a custom connector takes a few minutes: paste the server URL, sign in to CVPitch and approve the permissions. An owner may need to allow custom connectors in business plans of Claude or ChatGPT first.

Bottom line

MCP lets your assistant work with your recruitment software without copying and pasting. Connect CVPitch at https://cvpitch.app/mcp, start with metadata, share candidate content only on purpose, and remember that approval always stays with a person in CVPitch.