Skip to content

Building a ChatFrom chatbot from an AI agent

Updated

Building and running ChatFrom chatbots from Claude Code, Cursor or another AI agent, through ChatFrom's MCP server and an API key.

Can an AI agent build my chatbot?

Yes. ChatFrom has an MCP server, the standard way AI agents use other services. An agent such as Claude Code or Cursor can create a chatbot, add your website to its knowledge, write its instructions, try it and put it live, from a request in plain words such as "make a chatbot for my site from its help pages and put it live there". It goes through the same checks as the dashboard, and it works as you: it sees your chatbots and nobody else's, and your plan's limits and credits apply as they do to you.

How do I connect Claude Code?

  1. Open Account in the ChatFrom sidebar and find API keys for AI agents.
  2. Name the key after where it will be used, such as Claude Code on my laptop, and press Make a key.
  3. Copy the line the Account page shows with the new key and run it once in your terminal. Written out, with your key in place of cf_your_key, it is:
claude mcp add --scope user --transport http chatfrom https://chatfrom.io/api/agent/mcp --header "Authorization: Bearer cf_your_key"

--scope user makes ChatFrom available in every folder you open Claude Code in. To check it, run claude mcp list: ChatFrom should show as connected. A session that was already open does not see it, so start a new one. A good first request is "list my ChatFrom chatbots".

How do I connect Cursor or another agent?

In Cursor, add ChatFrom to ~/.cursor/mcp.json, or to .cursor/mcp.json in one project:

{
  "mcpServers": {
    "chatfrom": {
      "url": "https://chatfrom.io/api/agent/mcp",
      "headers": { "Authorization": "Bearer cf_your_key" }
    }
  }
}

Any other agent that connects to an MCP server over HTTP works the same way: the address is https://chatfrom.io/api/agent/mcp, and the key goes in the Authorization header as Bearer and the key.

How do I connect from Claude, without a key?

Claude's connectors sign you in instead of asking for a key. Open ChatFrom in Claude's directory at https://claude.ai/directory/chatfrom, add it and press Connect. A custom connector with the address https://chatfrom.io/api/agent/mcp does the same. ChatFrom asks you to sign in and shows what Claude will be able to do. Press Allow and you are back in Claude.

To stop it, open Account in the ChatFrom sidebar, find Connected apps and press Disconnect. It stops at once.

If an agent you gave a key to asks you to sign in instead, the key no longer works. Make a new one under Account.

What can the agent do?

  • list_chatbots, get_chatbot: your chatbots, and one chatbot's instructions, settings, embed snippet and share link.
  • create_chatbot: a new chatbot, as a draft only you can talk to.
  • update_chatbot: its name, instructions, model, language, welcome message, suggested questions, fallback message, tone and detail; its guardrails; memory; and whether it hands a conversation to a person, and to your own email or one you already confirmed.
  • find_pages, add_pages: finds a website's pages, then adds the ones it picks to the knowledge.
  • add_text, add_qa, list_knowledge, remove_source: texts and questions with their answers, the list of sources, and removing one.
  • ask_chatbot: asks the chatbot a question, as you would in Try it, and reads its answer and sources.
  • deploy_chatbot: puts the chatbot live or back to a draft, sets the websites it may run on and how the widget looks, and gives the snippet and share link.
  • list_conversations, get_conversation: what visitors asked, filtered to the unanswered, the rated down or the handed off, and one conversation in full.
  • list_evaluations, add_evaluation_dataset, run_evaluation: test questions with what a good answer does, and a scored run against them, billed to your OpenRouter key.
  • get_usage: the credits used this month, and when they start again.

Can I use ChatFrom from n8n or my browser?

Yes, with the same API keys. The ChatFrom node for n8n lets a workflow ask your chatbots, add pages, texts and Q&As to their knowledge, read their conversations and check your credits, and it works as a tool for n8n's AI Agent: install n8n-nodes-chatfrom under Settings, Community nodes, as chatfrom.io/docs/n8n shows. The ChatFrom Web Clipper for Chrome adds the page you are on to a chatbot in one click: see chatfrom.io/docs/extension.

What can the agent not do?

  • Delete a chatbot or your account, or change your plan or billing. Those stay in the dashboard, where they ask you to confirm.
  • Make, see or delete API keys, so a key cannot make another key.
  • Connect chat apps or services such as WhatsApp or Shopify. You set those up in the dashboard.
  • Send hand-offs to a new email address. It can set the address you sign in with, or one already confirmed for another of your chatbots; any other you set under Behaviour, which sends that address a link to confirm.
  • Go past your plan: a chatbot, a source or a credit over its limit is refused, with the reason, as it would be for you.

Does it cost anything?

The MCP server and API keys are included on every plan, the free Hobby plan too. Answers from ask_chatbot and adding knowledge use your monthly credits, the same as in the dashboard. Each key can make 120 requests a minute, more than an agent needs.

How do I keep my key safe?

A key starts with cf_ and is shown once: ChatFrom keeps only a fingerprint of it. It works as you, so treat it like a password: keep it in your agent's settings, never in code you share or a public repository. An account can have 10 keys; one for each agent or computer lets you stop one without stopping the others. The Account page shows when each key was last used, and deleting a key stops it at once.

Conversations hold what your visitors wrote. When your agent reads them, their words go to your agent and to the company whose AI model it runs on, and a visitor can write words meant to steer it. ChatFrom's tools tell your agent to read them as information, never as instructions; read what it proposes before it acts on a conversation.