MCP Server for Your AI Twin: How ChatGPT and Claude Can Ask Your Twin Questions

The Model Context Protocol (MCP) is an open standard that lets AI applications connect to outside systems such as files, databases and business tools. Anthropic introduced it on November 25, 2024 (Anthropic, 2024), and the official documentation compares MCP to "a USB-C port for AI applications" (Model Context Protocol, 2026). With MCP, a source builds one server and every compatible assistant can use it, instead of a custom integration for every app.

Claude's custom connectors and ChatGPT's custom MCP servers both take a web address rather than a program on your computer, so a twin hosted online works with both (Claude Help Center, 2026; OpenAI, 2026). The protocol also keeps changing: the July 28, 2026 revision removed protocol-level sessions and the opening handshake, so every request now carries its own protocol version and client capabilities (Model Context Protocol, 2026).

Anthropic created MCP, but the protocol no longer belongs to Anthropic. On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, Amazon Web Services, Cloudflare and Bloomberg (Anthropic, 2025). At the handover, Anthropic counted more than 10,000 active public MCP servers and more than 97 million monthly SDK downloads across Python and TypeScript (Anthropic, 2025).

Our verdict: with OpenAI, Google and Microsoft all backing one foundation, MCP is the safe bet for reaching AI assistants in 2026. Your clients do not all use the same assistant, and one MCP server reaches every assistant that adds it.

An MCP server is only an address that answers requests in a standard format, and nothing in the protocol says the answers must come from a database. Most servers wrap software such as a calendar, but a professional's knowledge is a source too. For someone deciding whether to hire you, your answer about fit, process and fees beats a summary of your website. A personal AI is an AI built around one person, and a personal MCP server is the door that lets other AIs consult that person. Without that door, your twin answers on your website and in apps connected through the API; with it, any assistant that connects your twin gets your answer in the first person, labeled as coming from your twin.

Timeless AI™ Twin's MCP door gives an assistant two tools, ask and about, and both only read. The ask tool puts a question to your twin, who answers in the first person from what you chose to share, and the assistant can ask follow-ups in the same conversation. The about tool returns your public introduction: who you are, what you offer and how to reach you. For example, here is how an exchange with a real estate agent's twin runs.

1. A home buyer using Claude, with the agent's twin connected, asks whether the agent works with first-time buyers in Phoenix, and Claude calls ask.

2. The twin answers in the first person, in the agent's own words, and points to the agent's booking link.

3. Claude shows the answer and can follow up in the same conversation.

The door cannot edit your twin, read your private settings or act for you. When someone wants to meet, your twin sends them to your booking link. Over MCP your twin answers in text, while replies in your own cloned voice happen on your website when a visitor speaks.

Any assistant or agent that can add a remote MCP server can connect, which as of October 2026 includes Claude, ChatGPT and the main developer tools.

Our pick for a first test is Claude, because its custom connector form takes your twin's address plus a header for your key, and the Free plan gets one custom connector (Claude Help Center, 2026). ChatGPT uses the same standard, but OpenAI's documentation says ChatGPT cannot present custom API keys, so your key travels inside the address instead (OpenAI, 2026). Gemini CLI accepts custom headers as well as OAuth (Google, 2026).

You need two things from your Timeless AI™ Twin settings: the address for AI assistants and the door's secret key. The key is shown once, when you create the door, so copy it somewhere safe. Every door is covered in the Timeless AI™ Twin guide.

Get your twin's address and key

1. Build your Persona free with 25 memories and no card, then move to Timeless Pro, which adds Timeless AI™ Twin.

2. In your twin settings, add your booking link and the questions clients ask most, so your twin answers them your way.

3. Open the door for AI assistants and copy its address and secret key.

Add your twin to Claude

1. In Claude on the web or Desktop, go to Customize, then Connectors.

2. Click + Add, then Add custom connector.

3. Enter a name people will recognize, such as your name plus "twin", and paste the address.

4. Under Authentication, choose No sign in.

5. Under Request headers, add Authorization with the value Bearer, a space and your key, then save.

6. In a chat, open the + menu, choose Connectors and switch your twin on.

Add your twin to ChatGPT

1. Open ChatGPT's Plugins page and select the plus button.

2. Choose Add custom MCP server, then enter a name and a short description.

3. Under Connection, paste your twin's address, then a slash and your secret key, because ChatGPT cannot send the key in a header (OpenAI, 2026).

4. Set authentication to none, since the key in the address does that job, then review the risk warning and select I understand and want to continue.

5. Select Create as a plugin, and check that ChatGPT found the ask and about tools.

That full address now works as a password, so keep it private, and if it leaks, remove the door and open a new one. For developer tools, your twin settings also show a ready-made settings block in the JSON format most MCP clients take. Whichever assistant you use, test with a question a real client asks, such as what your first consultation includes.

Yes, if you treat the key like a password and teach your twin only what you would tell a prospective client. The best-known MCP attack runs the other way: a malicious server hides instructions in its tool descriptions, which Invariant Labs named tool poisoning in April 2025 (Invariant Labs, 2025). That is why Anthropic tells Claude users to connect only servers they trust (Claude Help Center, 2026).

• Read-only by design. Both tools only answer and describe, and no tool can change your twin, your Persona or anything else in your account.

• Your words only. Your twin answers from what you chose to share, and the about tool returns your public introduction, never your private instructions.

• A key for this door. The MCP door has its own secret key, and you can pause or remove the door without touching your website or API doors.

• Named as your twin. The tools are labeled as your Timeless AI™ Twin, and asked directly, your twin says in one line that the person is talking to your twin.

• Assistant controls still apply. Claude asks you to approve tool calls, advises choosing Allow always only for servers you trust, and lets you switch off tools you do not need (Claude Help Center, 2026).

• Your succession rules. Under Executor Lock™, each twin door starts switched off after a verified death unless you chose in advance to keep that door answering, and your twin passes to your executor.

Use it wherever someone else's AI needs your actual answer. The best fits are people whose clients ask the same questions before they buy, such as consultants, coaches, financial advisors and real estate agents.

• Client research. A prospective client who connects your twin can ask about your services, fit and process inside the assistant they already use, then follow your booking link.

• Team answers. Staff who work in Claude or ChatGPT can ask the founder's twin how the founder handles onboarding questions, which keeps answers consistent; see our AI knowledge base for your business guide.

• Course questions. A professor's twin can answer syllabus and office-hours questions from inside a student's assistant, in the professor's own words.

• Quality checks. Connect your twin to your own assistant and ask the hard questions clients ask, so you see exactly what they will see.

Build your own if you are a developer who wants your assistant to query your own files. Use a twin if you want other people's assistants to hear your answers without you running a server.

A homemade server is a fine weekend project, but every spec revision, like the one in July 2026, lands on your desk. A custom GPT of yourself is quick to make, but only reaches people inside ChatGPT. Delphi lists API access only on its White-Glove Set Up plan, and its pricing page does not mention MCP as of October 2026 (Delphi, 2026), one reason people look for a Delphi alternative. Our verdict: for professionals, a twin with an MCP door wins, because one source of truth answers on your website, in your apps and in every assistant that connects. For a whole firm, see how to build a digital twin for your business, and on lock-in, read will your AI survive model changes.

Building your Persona is free, and the MCP door comes with Timeless AI™ Twin on Timeless Pro at US$59.99 a month or US$599 a year. The free build includes 25 memories, up to 5 Trusted Contacts and Executor Lock™ setup, with no card, and never expires.

Worked example (illustrative numbers). A year of Timeless Pro paid monthly costs 12 × $59.99 = $719.88, so the $599 annual plan saves $120.88. Say you spend 10 minutes each on 12 introductory questions a week, which is 120 minutes a week, or 104 hours over 52 weeks. If your twin handles half of them on your website and in the assistants your clients connect, you get back 52 hours a year, worth $7,800 at a $150 hourly rate, against a $599 plan. Full plan details are on the pricing page.

• Anthropic: Introducing the Model Context Protocol

• Anthropic: Donating the Model Context Protocol and establishing the Agentic AI Foundation

• Model Context Protocol: What is the Model Context Protocol (MCP)?

• Model Context Protocol: Specification key changes, 2026-07-28

• Claude Help Center: Get started with custom connectors using remote MCP

• OpenAI Apps SDK: Connect from ChatGPT

• OpenAI Apps SDK: Authentication

• TechCrunch: OpenAI adopts rival Anthropic's standard for connecting AI models to data

• TechCrunch: Google says it'll embrace Anthropic's standard for connecting AI models to data

• Microsoft: Model Context Protocol (MCP) is now generally available in Microsoft Copilot Studio

• Google: MCP servers with the Gemini CLI

• Invariant Labs: MCP Security Notification: Tool Poisoning Attacks

• Delphi: Pricing