Agent Quickstart

Point an AI coding agent — GitHub Copilot, Cursor, Claude Code, Windsurf, or anything that speaks MCP — at Chatterfly, and have it author, validate, and deploy workflows for you conversationally. This page covers the one-time setup, then gives you ready-to-paste prompts.

Before you start

You'll need:

A Chatterfly account — sign up free if you don't have one.
A coding agent that supports MCP (GitHub Copilot, Cursor, Claude Code, Windsurf, etc.), or a terminal if you'd rather have the agent shell out to the CLI.

Get a personal access token

Create a token

In the dashboard, go to Settings → API tokens and create one. It looks like cfpat_… and is shown only once — copy it somewhere safe.

Warning: Treat this token like a password — it authenticates as you, scoped to this workspace. Revoke it anytime from the same page.

Configure your agent (MCP)

Add the Chatterfly MCP server

Settings → API tokens has a generator that fills in your real token for VS Code, Cursor, Claude Code, or Windsurf — copy the config it produces and save it to the file it names. The shapes are shown below for reference.

Cursor and Windsurf — save as .cursor/mcp.json or ~/.codeium/windsurf/mcp_config.json respectively:

json

{
  "mcpServers": {
    "chatterfly": {
      "url": "https://api.chatterfly.in/mcp",
      "headers": {
        "Authorization": "Bearer ${CHATTERFLY_TOKEN}"
      }
    }
  }
}

Claude Code — save as .mcp.json for project scope or add it to ~/.claude.json for user scope. Claude Code requires the explicit HTTP transport type:

json

{
  "mcpServers": {
    "chatterfly": {
      "type": "http",
      "url": "https://api.chatterfly.in/mcp",
      "headers": {
        "Authorization": "Bearer ${CHATTERFLY_TOKEN}"
      }
    }
  }
}

VS Code (Copilot) — a slightly different shape, saved as .vscode/mcp.json:

json

{
  "servers": {
    "chatterfly": {
      "type": "http",
      "url": "https://api.chatterfly.in/mcp",
      "headers": {
        "Authorization": "Bearer ${CHATTERFLY_TOKEN}"
      }
    }
  }
}

Replace ${CHATTERFLY_TOKEN} with the token from the previous step. See For coding agents for the full tool reference.

Optional: also install the CLI

Install and log in

Some agents work better shelling out to a command than calling MCP tools directly — install the CLI so yours has that option too:

bash

npm install --global @chatterfly/cli
chatterfly login --token cfpat_your_token_here

See the CLI docs for the full command reference.

Example prompts

Paste these directly into your agent once it's connected. Swap in your own details where noted.

Build a workflow from a description

Load the Chatterfly DSL reference (get_dsl_reference: core, nodes),
      then write workflows/customer-feedback.json as a two-step workflow: a Message node
greeting the customer, then an Input node asking for a 1-5
      satisfaction rating. Treat that file as the source of truth. Validate
      the exact file, fix any errors, then create and deploy it with public
      access enabled so I can embed it on a web page. Finally, fetch the
      deployment snapshot and confirm that its definition matches the file.

Validate a local file

Read workflow.json in this repo and validate it against the
Chatterfly DSL. Fix any errors it reports and keep re-validating
until it passes, explaining each fix you make.

Wire up a tool-calling agent

List my Chatterfly connections and available tools, then add an
Agent node to my workflow that can call whichever ones are related
to sending email. Validate before saving.

Deploy and get a stable integration key

Deploy my Chatterfly workflow named "<workflow name>" and create a
stable workflow key for it so my backend can start runs without
the key breaking on the next redeploy.

Where to go next

For coding agents — the full MCP tool reference, llms.txt, and AGENTS.md.
CLI — every command, flags, and CI usage.
Workflow DSL — node types, branching, and validation rules your agent will use.
Prefer building visually instead? See the Quickstart.