The browser does not open

Start the connection from your assistant with Connect, Authenticate or Sign in. Opening the MCP address itself in a browser does not start the connection.

My assistant asks for a token or API key

Stop and never paste a credential. Use the Connect or Authenticate action of your assistant. Chainfill only works with signing in through the browser.

Chainfill says new connections are unavailable

AI connections are not switched on for your team yet. Ask your Chainfill contact to switch them on, or email support@chainfill.ai.

The assistant cannot change anything

You chose Read only, or your role does not allow the change. Revoke the connection and connect again with Read and edit, or ask an admin about your role.

I am viewing Chainfill as someone else

You cannot connect while you view the app as another user. Go back to your own account and try again.

My partner account has no customers yet

No customer is linked to your partner account yet. Ask Chainfill to link one.

I want to send a file from my computer

That does not work over the connection. Upload the file in Chainfill itself.

MCP server

Connect Chainfill to your AI assistant

Work in Chainfill from the AI assistant you already use: ChatGPT, Claude, Claude Code, Codex, Cursor, VS Code with GitHub Copilot or Gemini CLI.

MCP address

https://api.chainfill.ai/mcp

You sign in through the browser with your own Chainfill account. You need no token or API key.

Steps per assistant

Let your agent set it up

Working with an assistant that can run commands itself, such as Claude Code, Codex, Gemini CLI or Cursor? Paste this prompt. The assistant sets up the connection. You approve it yourself in the browser.

Prompt for your assistant

Connect Chainfill to this assistant. Follow the steps at https://www.chainfill.ai/en/mcp. Sign in through the browser and do not ask me for a token or API key.

In ChatGPT and Claude you add the connector yourself. Pick your assistant below.

For agents: these steps are also available as Markdown, and llms.txt describes everything Chainfill offers agents.

Steps per assistant

Every assistant uses the same address. Labels can differ by plan; use the labels your assistant shows.

ChatGPT

With Developer mode

Turn on Developer mode in Settings > Security and login.
Open ChatGPT Plugins, select + and enter the MCP address under Connection.
Create the connection and approve in Chainfill when ChatGPT asks.

Note: Your workspace can restrict Developer mode and custom plugins. If so, ask your ChatGPT administrator for permission.

Claude Code

Plugin or server only

1
Install the Chainfill plugin. It adds the connection and the Chainfill skills in one go. At the second command, choose Install for you.
Claude Code
```
/plugin marketplace add https://www.chainfill.ai/claude-plugins/marketplace.json
/plugin install chainfill@chainfill
```
Run /mcp in Claude Code, select Chainfill and sign in through the browser.

Or this way

Only want the connection, without skills? Add the server in your terminal instead and continue with step 2.

Terminal

claude mcp add --transport http --scope user chainfill https://api.chainfill.ai/mcp

Codex

Codex CLI and IDE extension

1
Add Chainfill and start the sign-in. The second command opens Chainfill in your browser.
Terminal
```
codex mcp add chainfill --url https://api.chainfill.ai/mcp
codex mcp login chainfill
```
Sign in and approve the connection.

Or this way

In the Codex IDE extension, choose MCP servers > Add server > Streamable HTTP, save the MCP address and select Authenticate.

Cursor

One click or mcp.json

Add to Cursor

Install in one click, or follow the steps below.

1
Put only the address in your personal ~/.cursor/mcp.json. Do not add an auth or headers block.
~/.cursor/mcp.json
```
{
  "mcpServers": {
    "chainfill": {
      "url": "https://api.chainfill.ai/mcp"
    }
  }
}
```
Start Chainfill in the MCP settings of Cursor and sign in through the browser.

VS Code

With GitHub Copilot

Install in VS Code

Install in one click, or follow the steps below.

Run MCP: Add Server, choose HTTP, enter the MCP address and choose Global.
Start Chainfill from MCP: List Servers, trust the server and sign in through the browser.

Or this way

Or add the server to your user configuration with MCP: Open User Configuration.

mcp.json

{
  "servers": {
    "chainfill": {
      "type": "http",
      "url": "https://api.chainfill.ai/mcp"
    }
  }
}

Gemini CLI

Add Chainfill as an HTTP server.

Terminal

gemini mcp add --transport http --scope user chainfill https://api.chainfill.ai/mcp

Gemini starts the sign-in itself. If the browser does not open, run /mcp auth chainfill in Gemini CLI.

Then, in your browser

You sign in to Chainfill as usual: with password, two-factor authentication or SSO.
If you have more than one team, you choose which teams the assistant may use.
You choose the access level and then Approve connection.
The browser sends you back to your assistant. You copy nothing.

Does it work? Ask your assistant, for example: "Which document types are available in Chainfill?"

What you can ask your assistant

Ask in your own words. The assistant picks the Chainfill tool that fits.

“Why did order TO-4500456120 not arrive in our TMS?”
The assistant looks up the document and explains what went wrong, with the error message and the status.
Tool:document.why-failed
“Why does Transport Veldhoven go wrong more often than other customers?”
The assistant compares the figures per customer, checks which fields people keep correcting and proposes a better customer note. You approve it.
Tool:analytics.supplier-performancecustomer-memory.propose
“Approve the orders that are ready.”
The assistant looks up the orders and approves them once you agree.
Tool:document.approve
“Resend the failed export of order TO-4500456120.”
The assistant sends the document to your system again and shows you where it stands.
Tool:document.resend
“Add a rule for Transport Veldhoven: the unloading date is always under Delivery.”
The assistant writes the rule for that customer and shows it to you before saving it.
Tool:supplier-rule.create
“How many documents did we process this month, and how many of them automatically?”
The assistant fetches your company overview: the number of documents, the accuracy, how many went through automatically and the weekly trend.
Tool:analytics.business-overview

Safe from the start

The assistant only sees what your Chainfill role allows, in each team it works in. The access level never grants more than your role.

You choose what the assistant may do

Read only
Look up documents, settings and data. Nothing is changed.
Read and edit
You approve each change. Assistants that support it show a confirmation form; others are told to check with you first.
Read and edit without asking
Changes run straight away, for example for automations.

You can revoke a connection at any time in Settings > Connected AI clients. Access stops on the next request.
The assistant never gets your password or stored credentials. Signing in and consenting happen on the Chainfill website.
Partners connect all their customer teams in one go, including teams added later.

If it does not work

Is your question not listed? Read the knowledge base article (sign-in required) or email support@chainfill.ai.

Connect Chainfill to your AI assistant

Let your agent set it up

Steps per assistant

ChatGPT

Claude

Claude Code

Codex

Cursor

VS Code

Gemini CLI

Then, in your browser

What you can ask your assistant

Safe from the start

You choose what the assistant may do

If it does not work

Ready to remove the manual work?