MCP is a standard that lets Claude, Codex or Cursor use MyChatBot as a set of tools. This page explains what is exposed, how to connect, and what stays in your hands.
Generate media and speech from your AI assistant; generation is billed to your account balance and needs no particular plan.
Manage assistants, leads, channels, follow-ups and the knowledge base, and read chats, orders and calendar events, through more than 100 Sales-Management tools.
Create and schedule social posts and read post analytics; this needs a paid social plan, and the docs do not list which networks are supported.
List ad accounts, ads and campaigns on the Scale plan or higher; creating or budgeting campaigns is not documented. Routing Meta lead forms to a sales assistant is a separate Sales-Management feature with its own plan conditions.
MCP, the Model Context Protocol, is a standard that lets an AI assistant use tools outside the chat window: it asks a tool to do something, gets an answer and carries on.
In MyChatBot’s documentation, MCP means endpoints that an outside AI client can connect to. MyChatBot calls each endpoint a surface, and the AI app on your side (Claude Code, Codex, Cursor or the Claude app) is the client. You describe a job in ordinary language and the client uses the surface’s tools.
Work you want to repeat is saved as a routine, a multi-step workflow stored in MyChatBot that runs there. Claude or Codex does not stay in the execution path.
MCP covers two different set-ups, and these pages are mostly about the first.
For the second direction, see the docs’ connectors cheat sheet and our MCP integration guide for sales assistants. The connector form asks you to acknowledge that custom connectors can introduce risks.
MyChatBot exposes five MCP surfaces, and the docs advise choosing the narrowest one that fits the job (see the surface map).
Build and test agents, skills, connectors, knowledge, routines, schedules and triggers. It can validate, save and preview a routine, but it cannot start an immediate live run or read customer records.
Manage assistants, leads, channels, follow-ups and knowledge base; read chats, orders and calendar. It has more than 100 tools, grouped into areas you can attach one at a time.
Search one product catalog. Every tool on this surface is read-only.
Generate media and speech, and use posting, post analytics and ads tools, each as its own area with its own requirements.
?domain=content, posting, analytics or ads.Search and read MyChatBot’s own documentation, so your AI assistant can look up exact feature and button names.
Claude Code and Codex have two routes, and the Claude app has its own connector.
config.toml block. The full key is shown only once.No terminal is needed. Add MyChatBot as a connector from the Claude app’s directory, sign in with your email and a one-time code, and tell Claude what you sell, as in the setup guide. The docs describe it only for building a sales assistant. Adding it by hand as a custom connector needs a paid Claude plan, the docs note.
claude mcp list shows registered servers.The docs set these boundaries, and your client does the reviewing.
Your client is the last gate: it can be set to auto-approve writes, so keep it on a setting that asks. In Codex, the documented writes mode asks before changes and billed previews.
Some limits are stated in the docs, and some details are not published.
Next: content, sales, publishing or advertising.
MCP, the Model Context Protocol, is a standard that lets an AI assistant such as Claude or Codex use outside tools. In MyChatBot it lets that assistant configure, query or learn about the platform through five surfaces: Agents, Sales-Management, Product, UGC and Docs.
No, you describe the job in ordinary language and the client calls the tools. With the Claude app no terminal is needed. With Claude Code or Codex you run one command or paste one config block that MyChatBot gives you, or sign in through the plugin in your browser.
The docs set boundaries rather than promise safety in general. Write tools and billed previews need explicit review in your client, schedules and triggers created over MCP start disabled, and publishing needs separate approval in the plugin flow. Keys and Product URLs are credentials, and a routine that has started runs to the end without a pause. Keep keys private, connect one area at a time and review what the client proposes.
The docs name Claude Code, Codex, Cursor and the Claude app, and say any compatible MCP client can connect. Claude Code and Codex also have a plugin with browser sign-in, so you do not copy a key.
The documentation does not list a separate charge, plan or rate limit for MCP access itself. What can cost money is the work you ask for. For example, content generation bills your account balance, a routine preview is billed from the Agents balance, and replies that sales assistants send, including follow-ups and outreach, use your sales message balance. Some areas have their own requirements: posting and post analytics need a paid social plan, and listing ads needs the Scale plan or higher. Check your plan and balance in the app.
That is the other direction. You add the outside server as a custom connector under Agents → Connectors, with its address, transport and any headers, and switch it on for the agents that need it. For a sales assistant it is the MCP tool on its Integrations page, which the guide MCP (Model Context Protocol) Integration at /guides/integrations-mcp walks through.
Create your account, connect your AI assistant to a single surface and ask it which tools it can see.