# Adding Connect to Claude

Add a custom connector pointing at `https://connectbyjbrh.com/mcp` with authentication set to none. There is no sign-in step and no consent screen, because the server exposes only public documentation. A successful connection shows ten read-only tools; if you see none, the connection did not complete.

- **Status:** Available
- **Audience:** developer
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/developers/mcp-quickstart-claude/

## Setting it up

1. Open the connector settings in your Claude client and choose to add a custom or remote MCP server.
   - Result: You are asked for a server URL and, usually, an authentication choice.
2. Paste `https://connectbyjbrh.com/mcp` as the URL and set authentication to none.
   - Result: No browser window opens. There is no OAuth flow to complete here, so a client waiting for one has nothing to wait for.
3. Save, and let the client connect.
   - Result: The client issues its own handshake and lists the tools. Ten appear, all annotated read-only.
4. Ask a question that needs the documentation — for example, whether Connect can record a call.
   - Result: The model calls `search_public_docs` or `get_product_status` and answers with a page URL and the capability status word, rather than from memory.

> **Note** The tool descriptions themselves instruct a model to check status before making a claim. That is why an answer about a `foundation` capability should come back as *the model and service exist, the usable feature is narrower* rather than as a flat yes.

## What you can ask once it is connected

- What Connect does on a channel, and what it deliberately does not do.
- The status of a named capability, checked against the published manifest rather than inferred.
- An end-to-end workflow, stage by stage, including what fails at each stage.
- The canonical definition of a term, from the glossary.
- Which protocol versions Connect implements, with the version strings the machine artefacts carry.

What it cannot do is act. No tool on this server sends a message, books anything, or reads a workspace — see [Workspace MCP tools](/developers/mcp-workspace-tools/). If a model offers to send an email through Connect, it has invented the capability and the connector will not have a tool to match.

## When it does not connect

**No tools appear at all** — The handshake did not complete. Check the URL for a trailing path or a missing `https://`; the endpoint is exactly `/mcp` at `connectbyjbrh.com`.
**The client reports a 405** — Something sent a `GET`. This endpoint is POST-only, and the response body says so. A client built for an older MCP revision that opens a stream will hit this.
**The client reports a 403** — An `Origin` header was sent that the server does not recognise. Desktop clients normally send none; a browser-based client must be served from an allowed origin.
**Everything stops for a minute** — The rate limit is 120 requests per minute per address. Wait rather than retrying immediately — the window is a rolling sixty seconds.
**The model answers without calling a tool** — It is answering from memory. Ask it to check the documentation, or use the `check_capability` prompt, which exists precisely for that.

## What the connection actually costs

Nothing is stored on Connect's side: no session is minted, no account is created, and no record of your questions is kept in a workspace, because no workspace is involved. The server keeps a per-address request counter for rate limiting and nothing else.

On your side, a connector adds tool descriptions to every turn, so it costs context whether or not a tool is called. Ten tools with short descriptions is a small cost, but it is not zero, and it is a reason to connect it in the assistant where you actually ask about Connect rather than in every one.

## Questions

### Do I need a Connect account to use the connector?

No. The server is unauthenticated and exposes only what the website already publishes. Having a workspace changes nothing about what these tools return.

### Will it read my workspace if I am signed in on the website?

No. The connector holds no session and shares nothing with your browser. Even signed in elsewhere, the tools return the same public pages.

### Is there a desktop configuration file example?

The wire-level steps that any client performs, including a raw request you can send with `curl`, are on [Using Connect from any MCP client](/developers/mcp-client-generic/). Configuration file formats belong to the client, not to Connect.

## Related

- [The Connect MCP server](https://connectbyjbrh.com/developers/mcp-server/)
- [Public MCP tools](https://connectbyjbrh.com/developers/mcp-public-tools/)
- [Using Connect from any MCP client](https://connectbyjbrh.com/developers/mcp-client-generic/)
- [Adding Connect to ChatGPT](https://connectbyjbrh.com/developers/mcp-quickstart-chatgpt/)
- [MCP error shapes](https://connectbyjbrh.com/developers/mcp-errors/)
- [Model Context Protocol](https://connectbyjbrh.com/docs/protocols/mcp/)

## What this page is based on

- `backend/app/mcp_server.py` — the endpoint, tool registry and limits
- `webapp/.well-known/agent-card.json` — the published machine surfaces
- Connect capability registry (docs-source/facts.py)