Skip to content
Charming Docs
Esc
↑↓navigate↵open⌘Jpreview
On this page

Use Charming with your agent

Connect any chat agent or coding agent to Charming and build the same kind of app either way.

Add the connector

Most chat agents that support custom MCP connectors, the open standard agents use to add tools like Charming, take the same URL:

https://charm.ing/mcp

Open that client’s connector settings, paste the URL, name it Charming, and approve the one-time consent screen it shows. ChatGPT works differently: see ChatGPT below.

Coding agents like Cursor, Claude Code, and Codex can skip the connector and call the HTTP API directly. Both paths reach the same backend and produce the same apps.

Claude

Claude has full native support for custom connectors, on Free, Pro, Max, Team, and Enterprise. Free is capped at one connector. The flow is the same on claude.ai and Claude Desktop. Full walkthrough with screenshots: usecharming.com/setup/claude.

ChatGPT

Add Charming from the ChatGPT Apps directory listing and click Connect; no developer mode needed. Start a fresh chat, open the + menu, and enable Charming before you ask it to build. Full walkthrough: usecharming.com/connect/chatgpt.

Grok

Full native support on grok.com. Paste the connector URL above and Grok auto-discovers the tools. Full walkthrough: usecharming.com/setup/grok.

Gemini

Gemini CLI and Gemini Enterprise support custom MCP servers with the connector URL above. The consumer Gemini web and mobile app doesn’t support custom connectors. Full walkthrough: usecharming.com/setup/gemini.

Perplexity

Local MCP ships in the Perplexity macOS app. Remote custom connectors are available to Pro and Max subscribers. Full walkthrough: usecharming.com/setup/perplexity.

Coding agents

Cursor, Claude Code, Codex, Cline, Windsurf, Continue, Zed, Gemini CLI, and similar agents can wire up the MCP connector URL above, but calling the HTTP API directly is simpler and needs no install. Give the agent the endpoint; it sends an authorization token with each request. Codex users can instead install the public Charming plugin (codex plugin marketplace add https://github.com/tambo-labs/charming-codex-plugin.git), which bundles the MCP endpoint and starter workflow.

Technical

Adding a chat client’s connector is a client-side step you do in that client’s settings; Charming’s side of the job is completing OAuth (with Dynamic Client Registration) once the client redirects, so a connector needs no manual token. An HTTP-only agent instead asks Charming for a token itself, through the device-code pairing flow below.

Copy this prompt for your agent

Connect me to Charming and help me build my first app. Read
https://charm.ing/docs/guides/use-charming-with-your-agent.md first.
If you can make HTTP requests, ask me to approve the connection; if
you're a chat client, tell me which connector URL to add and ask me
to confirm once it's added. Then read
https://charm.ing/docs/build-mcp.md (or build-http.md for the HTTP
API) for the app skeleton.

How an agent performs this job

A chat client adds the connector itself once you paste the URL into its settings; from then on the MCP session carries your auth, so tool calls need no bearer token. An HTTP-only agent instead calls POST /api/pair/start to get a device code, shows you the printed user_code and the charm.ing/pair approval link, and polls POST /api/pair/poll until you approve, at which point it holds a bld_user_* token good for every app on your account. An agent that only needs one app can skip pairing and call POST /app directly: the first app needs no auth and its response includes a bld_app_* token scoped to that one app.

The contract

Full detail on the two token families, read-only vs. writing calls, and what a second agent can and can’t touch lives in Connect to any AI agent. This page only covers getting connected; that one covers what a connected agent can then do.

Was this page helpful?