--- title: "How to Build an MCP Server Without Writing Code" canonical_url: "https://chatthing.ai/blog/how-to-build-an-mcp-server-without-code" last_updated: "2026-11-10T21:45:31.013Z" meta: description: "How to build an MCP server without code: turn a Chat Thing agent into one that Claude, ChatGPT and Cursor can ask, with your knowledge and power-ups." "og:description": "How to build an MCP server without code: turn a Chat Thing agent into one that Claude, ChatGPT and Cursor can ask, with your knowledge and power-ups." "og:title": "How to build an MCP server without writing code - Blog - Chat Thing" "twitter:description": "How to build an MCP server without code: turn a Chat Thing agent into one that Claude, ChatGPT and Cursor can ask, with your knowledge and power-ups." "twitter:title": "How to build an MCP server without writing code - Blog - Chat Thing" --- [← Back to blog](https://chatthing.ai/blog)**Blog**# **How to build an MCP server without writing code** ![Zef](https://res.cloudinary.com/djyjvrw5u/image/upload/v1710941716/IMG_2278_d3b5e4fa69.jpg) Zef 10 Nov 2026 ~ 9 min read --- ![How to build an MCP server without writing code](https://res.cloudinary.com/djyjvrw5u/image/upload/v1794235300/hero_4eddd44698.png) --- ### **On this page** 1. [What is an MCP server?](#what-is-an-mcp-server) 2. [Two ways to build an MCP server](#two-ways-to-build-an-mcp-server) 3. [What your agent's MCP server offers](#what-your-agents-mcp-server-offers) 4. [How to create an MCP server from your agent](#how-to-create-an-mcp-server-from-your-agent) 5. [Connect an assistant](#connect-an-assistant) 6. [Check it works](#check-it-works) 7. [Limits and costs](#limits-and-costs) 8. [Moving from a Custom GPT?](#moving-from-a-custom-gpt) 9. [When to build your own instead](#when-to-build-your-own-instead) 10. [FAQ](#faq) 1. [Do I need to write any code?](#do-i-need-to-write-any-code) 2. [Which assistants can connect?](#which-assistants-can-connect) 3. [Which plans include it, and does it cost extra?](#which-plans-include-it-and-does-it-cost-extra) 4. [Can the agent do things, or only answer?](#can-the-agent-do-things-or-only-answer) 5. [Can I keep the server private?](#can-i-keep-the-server-private) 11. [Turn it on](#turn-it-on) Any Chat Thing agent can now be an MCP server. Turn on the **MCP server** card in your agent's **Channels** tab and you get a server URL that Claude, ChatGPT, Cursor, VS Code and Codex can connect to. From then on, those assistants can ask your agent questions and use its answers. It's on the Standard, Pro and Enterprise plans. We think it's the easiest way to make an MCP server, and it works a little differently from most. A typical MCP server wraps API endpoints, with tools like `get_order`, `search_docs` and `create_ticket`, and the assistant calling it works out which tools to use, in what order, and what the results mean. A Chat Thing MCP server puts a whole agent behind one `ask` tool. Your agent reads its instructions, searches its knowledge, uses its power-ups and sends back an answer, with links to the pages it used. ## What is an MCP server? MCP (the [Model Context Protocol](https://modelcontextprotocol.io)) is an open standard for connecting AI assistants to other software. An MCP server offers tools, and an assistant connected to it, such as Claude or Cursor, calls those tools in the middle of a conversation when it decides they'll help. Building one yourself usually means picking an SDK (Python, TypeScript, C# and others have one), writing each tool, hosting the server at a public address, and adding authentication and rate limits. Most guides on how to build an MCP server walk through exactly that. If you need a server that returns raw data from your own systems, they're the right place to start, and there's more on that [further down](#when-to-build-your-own-instead). ## Two ways to build an MCP server | | Write your own server | Turn on a Chat Thing agent's MCP server | | --- | --- | --- | | What you build | Code for each tool, plus hosting and authentication | An agent in the dashboard, then one card to turn on | | Tools the assistant sees | Usually one per endpoint or action | `ask`, and optionally `search` | | Who works out the answer | The calling assistant, from your tools' results | Your agent, then the calling assistant uses its answer | | Knowledge | Whatever your tools return | Your agent's data sources, such as your website, files and Notion | | Actions | Whatever you code | Your agent's power-ups, such as booking a meeting | | Where questions show up | Your own logs | **Conversations**, with **MCP** as the channel | Automation tools such as Zapier and n8n also let you create an MCP server without code. They expose actions or workflows as tools, which suits an assistant that should run a task in another app. A Chat Thing agent's MCP server answers questions from your knowledge, in your agent's voice, and can still use its power-ups. ## What your agent's MCP server offers Your agent's server has two tools. - **`ask`** asks your agent a question and returns its answer. Your agent answers with the same instructions, knowledge and power-ups it uses in your web chat and on your other channels. When its **Include sources for responses** setting is on, the answer comes with links to the web pages it used. Every answer also includes a `conversationId`, which the assistant passes back to ask a follow-up in the same conversation. - **`search`** is off until you turn on **Offer a search tool**. It returns passages from your agent's knowledge word for word, with the page or file each one came from. Search skips your agent's instructions, so only turn it on if everything in your agent's knowledge is fine to share as written. Power-ups work over MCP the same way they do in your web chat. If your agent can book meetings, for example, it can book one for the person using the assistant. ## How to create an MCP server from your agent You need to be a team owner or admin, on the Standard, Pro or Enterprise plan. The [MCP server docs](https://chatthing.ai/docs/channels/mcp) cover every setting. 1. Open your agent, select the **Channels** tab and turn on the **MCP server** card. The **MCP server settings** page opens. ![An agent's Channels tab in Chat Thing, with the MCP server card switched on next to the other channel cards](https://res.cloudinary.com/djyjvrw5u/image/upload/v1794328974/mcp_channel_card_9325c4b4c3.png) 1. Under **Access**, choose **Who can connect**. **Anyone with the URL** works in every assistant, including custom connectors in the Claude and ChatGPT apps. **Only clients with the secret key** asks each client to send a key: click **Generate secret** and copy the key straight away, because you can't see it again after saving. 2. Under **When should an assistant use this bot?**, describe what your agent knows about and what it can help with, for example "Answers questions about Acme's products, pricing and orders". Assistants read this to decide when to ask your agent. If you leave it empty, your agent's description is used. 3. Turn on **Offer a search tool** if you want assistants to be able to read your knowledge directly. 4. Click **Save**. 5. Copy the **Server URL**, or one of the setups under **Connect an assistant**. ![MCP server settings in Chat Thing: the Server URL with a copy button, Access set to Anyone with the URL, the When should an assistant use this bot? description, and the Offer a search tool switch](https://res.cloudinary.com/djyjvrw5u/image/upload/v1794328977/mcp_settings_25d6c0d2c2.png) Your server URL looks like this: ``` https://app.chatthing.ai/api/public/channels//mcp ``` ## Connect an assistant Under **Connect an assistant**, pick your assistant's tab. Each setup has your URL filled in and a copy button. - **Claude Code:** run the `claude mcp add` command in your terminal. - **Codex:** add the server to `~/.codex/config.toml`. - **Cursor:** click **Add to Cursor**, or add the server to `~/.cursor/mcp.json`. - **VS Code:** click **Add to VS Code**, or add the server to `.vscode/mcp.json`. - **Claude & ChatGPT apps:** add a custom connector in the app's settings and paste the server URL. For Claude Code, the command looks like this: ``` claude mcp add --transport http acme-support \ https://app.chatthing.ai/api/public/channels//mcp ``` ![The Connect an assistant panel in Chat Thing, with tabs for Claude Code, Codex, Cursor, VS Code and the Claude & ChatGPT apps, showing the claude mcp add command and a copy button](https://res.cloudinary.com/djyjvrw5u/image/upload/v1794328979/mcp_connect_assistant_74ee3658ad.png) The Claude and ChatGPT apps usually can't send a secret key, so use **Anyone with the URL** for them. The **Add to Cursor** and **Add to VS Code** buttons only appear with that option too, so a secret key never ends up in a link. With **Only clients with the secret key**, clients send the key as an `Authorization: Bearer` header, and the setups for Claude Code, Codex, Cursor and VS Code include it for you. Whether you can add a custom connector in ChatGPT depends on your ChatGPT plan and settings. If you can't find the option, check OpenAI's help centre for your plan. ## Check it works Ask your assistant something your agent knows, for example "Ask Acme Support how long delivery takes". The assistant calls your agent's `ask` tool and uses the answer. Then open **Conversations** in Chat Thing. The conversation is listed there with **MCP** as its channel, next to your web chats, so you can read what people asked from their assistants. ![Chat Thing's Conversations page with a conversation from an AI assistant, labelled with MCP as its channel](https://res.cloudinary.com/djyjvrw5u/image/upload/v1794328982/mcp_conversation_f9f5a2b6ab.png) ## Limits and costs - Every question uses your team's message tokens, like a question in your web chat. Standard is $49 a month and includes 19,000,000 message tokens a month, shared by all your agents ([compare plans](https://chatthing.ai/docs/account/plans)). [Message tokens](https://chatthing.ai/docs/account/message-tokens) explains how they're counted. - Each caller can ask up to 10 questions and run up to 30 searches a minute, and each agent answers up to 60 questions a minute in total. Over that, the assistant is told to wait a minute and try again. - Your agent's **Daily message limit per visitor** setting applies over MCP too. Callers are identified by their network address, and the limit is shared with your web chat. - Your agent's **Maximum characters per message** setting applies to questions. With **Anyone with the URL**, anyone who has the URL can ask your agent questions, and those questions use your tokens. Treat it like a link to your web chat, and set a **Daily message limit per visitor** before you post it anywhere public. The secret key is one key for everyone you give it to, and there's no sign-in per person. ## Moving from a Custom GPT? OpenAI has scheduled Custom GPTs to retire on 11 December 2026, and shared GPT links stop working then. If the people who used your GPT did so inside ChatGPT, a Chat Thing agent can meet them there: [move the GPT to Chat Thing](https://chatthing.ai/docs/get-started/move-from-a-custom-gpt), turn on its MCP server, and they can add it to ChatGPT as a custom connector. The same URL works in Claude and Cursor. There are two catches. Each person adds the connector themselves, so it doesn't replace a GPT link that anyone could open. And ChatGPT decides which of its plans can add custom connectors. For people who only ever followed your link, your agent's own page link or your website chat is the closer replacement. Our post on [Custom GPTs retiring](https://chatthing.ai/blog/custom-gpts-retiring) goes through the options. ## When to build your own instead Your agent's MCP server answers questions and runs your agent's power-ups. Write your own server when: - **Callers need structured data rather than an answer.** If another program needs `{"status": "shipped"}` and not a sentence, give it a tool that returns exactly that. - **Each person should only see their own data.** Your agent's server has one access setting for everyone, the open URL or a shared key. Per-person sign-in and document-level permissions need your own server with OAuth. - **You don't want a model in the middle.** Every `ask` runs a full agent turn, which takes longer than a database lookup and uses message tokens. - **It has to run on your own machine or network.** Your agent's server is hosted by us at a public URL. - **You want the assistant to drive lots of small tools itself.** If Claude or Cursor should do the reasoning across many endpoints, expose those endpoints directly. This is also separate from [Chat Thing's own MCP server](https://chatthing.ai/mcp), which lets you build, test and manage your agents from Claude, ChatGPT or Cursor. ## FAQ ### Do I need to write any code? No. You set up the agent in the Chat Thing dashboard, turn on the **MCP server** card and add the URL to your assistant. For Claude Code, that's one terminal command, and Chat Thing gives it to you ready to copy. ### Which assistants can connect? Claude Code, Codex, Cursor and VS Code have ready-made setups, and the Claude and ChatGPT apps connect through a custom connector. Other assistants that can add a remote MCP server by URL connect the same way. ### Which plans include it, and does it cost extra? Standard, Pro and Enterprise. There's no separate fee: each question uses your team's message tokens, like any other channel. ### Can the agent do things, or only answer? It can do whatever its power-ups do, such as booking a meeting, sending an email or calling your API. ### Can I keep the server private? Choose **Only clients with the secret key** and give the key only to the people who should connect. Connectors in the Claude and ChatGPT apps usually can't send a key, so they need **Anyone with the URL**. ## Turn it on If you're on Standard or above, open any agent's **Channels** tab and turn on **MCP server**. Every step and setting is in the [MCP server docs](https://chatthing.ai/docs/channels/mcp). New to Chat Thing? [Start free](https://app.chatthing.ai/app/auth/register) and build your agent, then move to [Standard](https://chatthing.ai/pricing) when you're ready to turn on its MCP server. ### **Related articles ** [![How to make a Telegram bot with BotFather](https://res.cloudinary.com/djyjvrw5u/image/upload/v1793391670/hero_58b5d93b1a.png)](https://chatthing.ai/blog/make-a-telegram-bot-with-botfather) [

How to make a Telegram bot with BotFather

](https://chatthing.ai/blog/make-a-telegram-bot-with-botfather) Step-by-step: create a Telegram bot with BotFather in about five minutes, with no code, then connect it to a Chat Thing agent so it can answer questions from your own content. [![Custom GPTs retire on 11 December: what changes and what to do](https://res.cloudinary.com/djyjvrw5u/image/upload/v1793633543/hero_a9b3f03945.png)](https://chatthing.ai/blog/custom-gpts-retiring) [

Custom GPTs retire on 11 December: what changes and what to do

](https://chatthing.ai/blog/custom-gpts-retiring) OpenAI is retiring Custom GPTs on 11 December 2026. What moves to plugins, what breaks for shared GPTs, and how to keep yours online. [![How to add an AI chatbot to your WordPress site](https://res.cloudinary.com/djyjvrw5u/image/upload/v1793305468/hero_9ef6db9cfa.png)](https://chatthing.ai/blog/add-ai-chatbot-to-wordpress) [

How to add an AI chatbot to your WordPress site

](https://chatthing.ai/blog/add-ai-chatbot-to-wordpress) Step-by-step: put a Chat Thing AI agent on any WordPress site with the free WPCode plugin, or embed it in a single page. Real screenshots from WordPress 7.1, plus fixes for the problems you're most likely to hit.
**Previous **
[← Home: see how all your agents are doing](https://chatthing.ai/blog/home-all-your-agents-on-one-page)