---
title: Connecting Your Coding Agent
---
# Connecting Your Coding Agent

You can connect a coding agent directly to your webhooks using [MCP](https://modelcontextprotocol.io).

## Connect your agent

<Steps>

### Open 'Settings' → 'MCP'

### Click 'Connect to MCP'

This generates an access token and opens a dialog with ready-to-paste setup steps. Pick your agent from the tabs and follow the steps there.

The token is only shown once, so copy it before closing the dialog.

### Check that your agent picked it up

Restart your agent if it was already running, then confirm the server is connected.

</Steps>

## What you can ask

Once connected, you can ask questions in plain language and let the agent do the digging:

- _"Why is the `invoice.paid` event failing?"_ — it searches the correct endpoint,  checks the endpoint's delivery stats, then the failed attempts and the exact response your handler returned.
- _"What does the payload for this event actually look like?"_ — it fetches a recent message and its JSON payload, so you can write your handler against the real thing.
- _"My handler was down for an hour — replay what it missed."_ — it recovers the endpoint from that point in time.
- _"Which endpoint is this message not reaching?"_ — it lists every endpoint the message was attempted against and how each one responded.


<Callout type="warning">

Resending, recovering, and updating a transformation are real actions: they send actual webhooks to your endpoints and change live configuration. Your agent should only do these when you ask it to, but review those tool calls before approving them, the same way you would any other agent action.

</Callout>

## Managing your token

Your tokens are listed in the **MCP** tab, redacted, with the date each was created and when it expires.

- **Tokens expire.** They're valid for 7 days by default.
- **Revoke a token whenever you want.** Open the token's menu and select **Expire**. Any agent using it loses access immediately.
- **Payloads reach your agent's model.** Anything the agent reads, including message payloads, is sent to the model provider behind your coding agent, so treat it the way you'd treat pasting a payload into that tool.