Why Build Your Chatbot From Claude Code or Codex

What does it mean to build a chatbot from Claude Code or Codex?

Claude Code and Codex are AI coding agents that work in your terminal or editor. Both can connect to outside services through the Model Context Protocol (MCP), a standard way for an AI tool to discover and call another service's tools.

IMFA Agents runs its own MCP server at https://site.imfa.app/mcp. Once you connect it, the coding agent you already use can do almost everything the AI chatbot builder does: create a chatbot, write its instructions, teach it from your website, design its chat widget, set up its workflow and channels, publish it and give you the embed code. Your agents, their conversations and their credits stay in IMFA Agents; the coding agent only drives them, as you.

When should you build your chatbot from a coding agent?

It works best when the chatbot sits next to other technical work:

• You are building the website anyway. The same session that writes the site can create the chatbot, style it to match and paste the embed snippet into the layout.
• You manage agents for several clients. One terminal reaches every agent you own or were invited to, instead of a builder tab per client.
• Your product changes often. When a feature ships, ask the agent to update the chatbot's instructions and knowledge in the same session.
• You like to describe, not click. Write what the chatbot should do in plain words and review what the agent built.

If you build one chatbot and rarely touch it, the visual builder is just as fast.

How to create an IMFA Agents chatbot with Claude Code or Codex

1. Connect the IMFA Agents MCP server. In Claude Code:

   claude mcp add --transport http imfa https://site.imfa.app/mcp

   Then run /mcp and choose Authenticate. In Codex, add the server to ~/.codex/config.toml and sign in:

   [mcp_servers.imfa]
   url = "https://site.imfa.app/mcp"

   codex mcp login imfa

   Both open an IMFA Agents page in your browser. Don't have an account yet? Sign up there and try every Pro feature free for 7 days.

2. Choose what the agent may do. On that page, pick Read only or Read and edit, whether to include private data (conversations, leads and contact details, off by default), and all your agents or only some. You can change or disconnect it any time on the MCP page. For scripts and CI, create an API key there instead.

3. Describe the chatbot. Write one clear request:

   Create an IMFA chatbot for "Atlas Bakery" in London. It answers questions about
   our menu, opening hours and custom cake orders, in French, Arabic and English.
   Friendly tone. Collect a name and phone number for cake orders. Hand off to a
   person for complaints. Don't publish yet.

   The agent calls agent_create and returns a draft with a link to open it in the builder. Nothing is public until you publish.

4. Teach it your facts. Ask it to add your website (knowledge_add_website, Pro) or write skills for detailed procedures (skill_save). Later, knowledge_list_gaps shows the questions visitors asked that the chatbot could not answer, and knowledge_answer_gap fixes them. Read the RAG guide for how answers stay grounded.

5. Design the widget. "Use our brand colors #8B4513 and #FFF8F0, put the launcher bottom left." The agent applies a preset or exact colors with design_apply_preset and design_set_colors. Design changes stay in the draft.

6. Set up the workflow. workflow_get shows what is connected. To add WhatsApp, Instagram, Telegram or Google Calendar, the agent calls connector_connect: you get the exact steps and a link, and you enter any bot token or key in IMFA Agents yourself. Secrets never pass through the AI.

7. Check, publish, embed. agent_get_readiness lists anything missing. When you ask to publish, agent_publish first returns a preview; the agent shows it to you and only publishes after you say yes. Then embed_get_snippet gives the one line of script to paste before </body>, or straight into the site it is building. See the website chatbot guide for the rest.

Stuck on a plan limit or out of credits? The IMFA Agents MCP answers with what is missing and a link to fix it, and account_get_status shows your plan, credits and next steps.

Why the IMFA Agents MCP makes this safe and fast

One place to work. The chatbot is built where you already write code and copy, without copying prompts between windows.

You stay in control. Publishing, pausing, deleting, inviting people and changing who may embed the chatbot always show you a preview first. After the agent reads something a visitor wrote, every change asks you first too, so a message cannot instruct your coding agent on its own.

Narrow permissions. Each connection gets only the access you chose. Read only cannot change anything, private data is off unless you turn it on, and you can limit a connection to one client's agents.

Everything can be undone. Every change made through the IMFA Agents MCP lands in the agent's history; agent_undo_last_change rolls it back, and the MCP page lists 30 days of activity.

The platform does the hard parts. Knowledge search, channels, lead capture, human handoff and safety rules run in IMFA Agents, so the coding agent works on what your customers need, not on infrastructure. To understand the protocol side, read why your AI chatbot needs MCP.

Frequently asked questions about building a chatbot with Claude Code or Codex

Do I need to know how to code?

No. You talk to Claude Code or Codex in plain words and they call the IMFA Agents MCP tools for you. Coding helps only if you also want the agent to edit your website.

Which plan do I need?

Every account starts with a free 7-day trial of Pro, card required, so you can create, design and publish a chatbot through MCP with website knowledge, other AI models, voice and one-click social channels from day one. After the trial, Pro is $20 a month. Without a plan your chatbots stop answering, and the MCP tells you what is missing.

Can the coding agent publish my chatbot without asking?

No. Publishing returns a preview first, and the change only happens after you confirm in your coding agent.

Can other AI tools connect to the IMFA Agents MCP?

Yes. Cursor, VS Code, Zed, ChatGPT and any MCP client can connect to https://site.imfa.app/mcp. The MCP page shows the setup for each one.

Does using the MCP cost credits?

Most building and editing does not. Reading website pages into knowledge, talking to a published chatbot with agent_chat, and the chatbot's real conversations with visitors use your credits as usual.