Model Context Protocol

Build and run your support agents from Claude Code, Cursor or any MCP client

Your whole support stack, controlled from wherever you work. Build agents, then summarise conversations, catch what they're missing and ship fixes. Connect in one command.

  • Secure OAuth connect
  • Scoped to your team
  • API keys for CI
or pick your client
claude mcp add chatthing --transport http https://app.chatthing.ai/mcp

Run this in your terminal. Your browser opens to approve access.

Endpoint: https://app.chatthing.ai/mcp

Works with your favourite AI agents

Set up with one prompt

Let your agent set itself up

Skip the config. Paste one instruction and your agent reads our docs, connects, and gets to work.

Set up ChatThing for me. Fetch https://chatthing.ai/docs/mcp-getting-started.md and follow the instructions.

Prefer to do it yourself? Read the getting-started guide →

Everything Chat Thing does, now agent-native

The same actions you'd take in the dashboard are exposed as MCP tools, so your agent can build and run a whole bot on your behalf.

Build & manage bots

Create bots, tune their model, prompt and retrieval settings, and update them - all without leaving your editor.

Add data sources

Point a bot at your website, docs or manual content. Discover pages, sync sources and check what's indexed.

Configure power-ups

Give bots real-world actions - look up orders, hit your API, hand off to a human - and toggle them on or off.

Set up channels

Configure the web channel and webhooks so your bot goes live on your site and wherever your customers are.

Test & evaluate

Write test cases, run test suites and read the results to catch regressions before your customers do.

Chat & inspect

Start chats, send messages and read conversations to verify behaviour end to end from your agent.

Things you could ask for

Once connected, just say what you want in plain language. A few starting points:

“Add our pricing page and changelog as data sources, then re-sync them.”

“Summarise my agent's last 50 conversations and pull out the top 3 things customers wanted.”

“Spin up a support bot from our help centre and put a talk-to-a-human escape hatch on it.”

“Find questions my bot answered badly last week and draft docs to fix the gaps.”

“Write test cases for our refund policy answers and run them against the bot.”

“Turn on enhanced retrieval and raise the context size for the support bot.”

From one prompt to a live bot

Connect over OAuth, then watch your agent build a support bot from your docs - the real MCP tool calls, start to finish.

agent — chatthing

Once connected, an agent runs the real Chat Thing tools - create_bot, discover_pages, add_data_source, update_bot, create_power_up, update_web_channel and start_chat - to build a working support bot from your docs in one prompt.

Prefer to script it? Generate an API key in your account and pass it as a bearer token - ideal for CI and headless agents.