Question 1

What is MCP (Model Context Protocol)?

Accepted Answer

MCP is an open standard that lets AI systems talk to external tools through one unified, secure interface. Instead of building a custom integration for every tool, you expose your data and capabilities once over MCP and any compatible AI client can use them. It is not specific to YourGPT. It is an industry standard, which is why the same YourGPT MCP server works with Claude Desktop, Cursor, ChatGPT, and others.

Question 2

What is an MCP server?

Accepted Answer

An MCP server is a service that exposes tools, data, or capabilities to AI clients over the Model Context Protocol. When you turn your YourGPT chatbot into an MCP server, it publishes your agent's knowledgebase behind a single secure endpoint, and any MCP-compatible client like Claude Desktop, Cursor, or ChatGPT can connect to it and query that knowledge without any custom integration work.

Question 3

What is the difference between MCP and an API?

Accepted Answer

A traditional API is a bespoke integration: every tool exposes its own endpoints, authentication, and data shapes, so an AI client needs custom code for each one. MCP is a standard layer on top of that. You expose your data once over MCP, and every compatible AI client discovers and calls it the same way, with no per-client integration. Think of MCP as a universal adapter for AI where APIs are the sockets — and YourGPT gives you that server without the plumbing.

Question 4

How does MCP work with YourGPT?

Accepted Answer

There are two distinct ways. First, Integrations turn YourGPT into an MCP server, letting you use your AI agent's knowledge and capabilities outside of YourGPT in tools like Claude Desktop, Cursor, and Windsurf. Second, Functions let your YourGPT agent use external MCP servers inside your conversations to fetch data or perform actions. Most people start with the first.

Question 5

What can I do with YourGPT's MCP server?

Accepted Answer

It gives AI assistants direct access to your chatbot as a knowledgebase. That means you can ask Claude or Cursor a question and get answers grounded in your own documentation, help articles, website content, and FAQs, the same knowledge your chatbot is trained on, without copying anything into the tool or leaving your editor.

Question 6

Which MCP clients does YourGPT support?

Accepted Answer

YourGPT works with any MCP-compatible client. The dashboard gives you ready-made configuration for Claude Desktop, Claude Web, ChatGPT, Coding Agents, CLI Agents, and OpenClaw. Because MCP is an open standard, clients released in the future will work with the same configuration.

Question 7

What is the difference between project-level and organization-level MCP access?

Accepted Answer

Project-level exposes one chatbot. You create an MCP server from that chatbot's Integrations page, and the client can only reach that agent's knowledge and nothing else. Organization-level works across your organization through a single API token created in Settings → API Tokens, scoped with an Access Role and a Project Scope of either all projects or specific ones. Use project-level to share one agent, and organization-level when you want one connection covering more than a single project.

Question 8

How do I connect my chatbot to Claude Desktop?

Accepted Answer

Go to Integrations in your YourGPT dashboard, find MCP (Model Context Protocol), click Configure, then Create MCP. Open the Claude Desktop tab, copy the configuration block, and paste it into your claude_desktop_config.json file — on macOS at ~/Library/Application Support/Claude/, on Windows at %APPDATA%\Claude\. Save the file and restart Claude Desktop.

Question 9

How do I connect my chatbot to Cursor or Windsurf?

Accepted Answer

The process is the same as Claude Desktop. Create your MCP server in the Integrations section, then open the tab for your tool. Each tab shows the exact configuration block and the file path where it needs to go for that specific tool. Copy it across, save, and restart the tool.

Question 10

Why isn't the hammer icon showing in Claude Desktop?

Accepted Answer

The hammer icon confirms Claude Desktop has loaded your MCP tools, so if it is missing the configuration has not been picked up. The usual causes, in order: the file is not valid JSON, and one stray comma is enough, so paste it into a JSON validator; you closed the window instead of fully quitting, so Claude never reloaded the config (use Quit from the dock on macOS or the system tray on Windows); or npx is not on the minimal PATH Claude Desktop launches with, which trips people up even when npx works fine in their terminal. Run 'which npx' and use the full path. Also check you edited the right file: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows. Claude writes a log per server named mcp-server-NAME.log if you need to dig further.

Question 11

Is my data secure when using MCP?

Accepted Answer

Every connection is authenticated with a bearer token, and a project-level MCP server is read-only, so a client can search your knowledgebase but cannot change anything. It covers that chatbot's knowledge, plus any connected projects you have chosen to share knowledge from. Treat the token like a password: anyone holding it can query that knowledgebase, so never commit it to a repository, paste it into a public issue, or share it in a screenshot. If it does leak, regenerate it from the dashboard to replace it.

Question 12

Can I revoke MCP access after granting it?

Accepted Answer

Yes, and the two levels work differently. For a single chatbot, Remove MCP on the Integrations page removes the integration, and regenerating replaces the token so the old one stops working. Organization-level API tokens give you finer control: each can be disabled or revoked individually from Settings → API Tokens, carry an optional expiry date, and record when they were last used, so you can retire one person's access without disrupting anyone else.

Question 13

Can I control what an API token can access?

Accepted Answer

Yes. When you create an organization-level API token you choose an Access Role (Owner, Editor, or Viewer) and a Project Scope, either all projects or specific ones. You can also enable fine-grained permissions to restrict which modules the token can reach, and set an expiration date. Note that only organization owners can create API tokens.

Question 14

Can each chatbot have its own MCP server?

Accepted Answer

Yes. Each chatbot can have its own MCP configuration, which lets you expose different agents for different use cases. A support agent and an internal documentation agent can each have their own MCP server and their own token, so you can share one without exposing the other.

Question 15

Do I need to reconfigure my MCP client after changing my chatbot?

Accepted Answer

No. The configuration points at your chatbot's MCP endpoint, not at a snapshot of its content, so retraining your agent or editing its knowledge sources does not require touching the config in Claude, Cursor, or any other client. You only need to update the configuration if you remove and regenerate the MCP server, which issues a new token.

Question 16

Can my YourGPT agent use external MCP servers?

Accepted Answer

Yes, through Functions. This is the reverse direction: instead of exposing your agent to other tools, your agent connects out to external MCP servers such as a database connector, file system access, or a custom utility, and uses them during conversations. You can point it at any MCP server, including ready-made ones from MCP360. MCP servers must be added via Functions, where they are treated as function calls the agent can invoke.

Question 17

How do I add an external MCP server to my agent?

Accepted Answer

Go to the Functions section, click MCP Function, and select Add your MCP Server. Configure the connection details, then make sure your agent's system prompt or persona is set up to allow function usage. Otherwise the agent will not call the tools you have added.

Question 18

What data sources can I expose over MCP?

Accepted Answer

Whatever your chatbot is trained on. That includes website links you have crawled, uploaded documents, FAQs and text snippets, and AI Helpdesk articles. If your agent can answer from it, an MCP client can reach it.

Question 19

Does MCP work with my existing chatbot, or do I need a new one?

Accepted Answer

It works with your existing chatbot. MCP exposes the agent you have already built and trained. There is nothing to rebuild or migrate. Open the chatbot you want to share, create its MCP server, and connect.

Question 20

Can my team use MCP across more than one project?

Accepted Answer

Yes. That is what organization-level access is for. A single API token works across your organization rather than a single chatbot, so you do not need to configure each project separately. Only organization owners can create tokens, and each should be issued to one person rather than shared: tokens are credentials, so a shared one makes it impossible to tell who did what, and revoking it cuts off everyone at once. Scope each token with an Access Role and Project Scope, and set an expiry.

Connect your AI Agent Anywhere with MCP

Simple MCP Integration Process

Go to Integrations & Configure MCP

Create your MCP Server

Select your Target Tool

Works With Your AI Tools

Claude Desktop

Cursor

Windsurf

Unlock use cases for your AI agent

Frequently Asked Questions