Explainer · MCP

Your proxy now answers inside ChatGPT and Claude

By Christopher Clement · October 17, 2026 · askproxy.ai

Every AskProxy proxy is also an MCP server. Add a proxy’s address, askproxy.ai/username/mcp, to Claude or ChatGPT, and the assistant can pull that person’s profile, ask their proxy questions, check their fit for a job and pass along a message, without leaving the chat. MCP, the Model Context Protocol, is the open standard that makes the connection.

There is nothing to switch on. If your proxy is live, it already works this way.

What MCP is

MCP is an open protocol for connecting AI assistants to outside tools and data. Anthropic introduced it in November 2024, and in December 2025 donated it to the Agentic AI Foundation, a Linux Foundation fund that Anthropic, Block and OpenAI co-founded.

An MCP server publishes a list of tools, each with a name, a description and the inputs it accepts. An MCP client, such as Claude or ChatGPT, reads that list and calls a tool when a conversation needs it, and the server sends back the result.

A proxy speaks MCP over Streamable HTTP, one of the two standard transports in the current specification, so any client that can reach a web address can use it.

The four tools a proxy exposes

• get_profile returns an overview of the person: role, summary, work history and how to contact them.
• ask puts a question to the person’s proxy, about their work, experience, projects or availability. It is the same proxy that answers at their own link.
• analyze_job_fit takes a public job listing URL or the job description text, and returns the alignments and honest gaps against the person’s real background.
• leave_message emails the person, with your name and email as the reply-to. Its description tells the assistant to use it only when you have asked to reach the person and given your real name and email, and never to invent contact details.

Two addresses, depending on who you ask about

• One person: https://askproxy.ai/username/mcp, for example askproxy.ai/christopher/mcp. The tools come pre-bound to that person, so the assistant never needs a username.
• Anyone with a proxy: https://askproxy.ai/mcp. Every tool takes a username, so one connection covers every proxy. This is the one to add if you look people up often, like a recruiter comparing candidates.

Connect it to Claude

Steps as of October 2026, from Claude’s help center.

1. Open Customize, then Connectors.
2. Click Add, then Add custom connector.
3. Give it a name, such as “Christopher’s proxy”, and paste the address.
4. Click Continue, choose No sign in, then Add.

Custom connectors work on every Claude plan, including Free, which allows one. On Team and Enterprise, an owner adds the connector for the organization first, and members then connect to it. Once it’s added, ask Claude about the person in plain language, and it calls the proxy’s tools when it needs them. A connection made on the web or desktop also shows up in Claude’s mobile apps.

Connect it to ChatGPT

Steps as of October 2026, from OpenAI’s help center.

1. Open Plugins from the sidebar and click +.
2. Choose Add custom MCP server, name it and paste the address.
3. Choose no authentication, accept the risk notice, and create it as a plugin.
4. Install it, then call it in a chat with @.

Availability is narrower than in Claude. Business, Enterprise and Edu workspaces can add custom MCP servers on the web, and on Business only admins and owners can create them. Pro accounts can connect them for read-only use. OpenAI’s article doesn’t list Plus, Go or Free.

ChatGPT treats any tool that isn’t marked read-only as a write action and asks you to confirm it. A proxy marks its three lookup tools read-only, so only leave_message, which sends an email, should ask.

Any other MCP client

Point it at either address over Streamable HTTP. There is no API key and no sign-in. A proxy is as public as its link.

Once connected, ask in plain language, the way you would ask a colleague: “What has she built recently?” or “Check his fit for this role,” followed by a job link. The assistant picks the right tool, so you don’t name tools yourself.

What stays the same inside an assistant

• The answers. Questions asked over MCP run through the same pipeline as questions asked at the person’s own link, with the same safeguards.
• The attribution. Every answer from ask and analyze_job_fit ends with a line naming it as the person’s proxy, an AI representative, not the person, with a link to their proxy.
• The limits. MCP questions count toward the owner’s monthly conversations, like questions asked on askproxy.ai.
• The leads. A message sent with leave_message reaches the owner by email, and their reply goes straight to the sender.

When someone asks an assistant about a person, the assistant works from whatever it can find. A proxy gives it a source built from the person’s own material, answering openly as their representative: about the person, never as them, whether the question comes from a browser tab or from someone else’s AI. More on why that matters: the digital twin wave has a trust problem.