> ## Documentation Index > Fetch the complete documentation index at: https://docs.signalmd.dev/llms.txt > Use this file to discover all available pages before exploring further. # Connect your client > Copy-paste setup for Claude Code, Claude Desktop, Cursor, VS Code and any other MCP client. Every client needs the same two things: the address `https://first-signal-api.fly.dev/mcp` and an API key sent as `Authorization: Bearer `. Only the file or command that holds them differs. [MCP overview](/mcp/overview) lists what the tools do. ## Before you start Open **API keys** in the app and select **New key**. Choose **Read** unless you want the assistant to create tests or start runs. [Authentication](/api/authentication#make-a-key) has the steps. The **Key made** dialog shows snippets for Claude Code, Cursor, Claude Desktop and curl with your key filled in. The snippets below cover the same clients and more, with `sig_your_key` as a stand-in. Swap in your own key, and do not commit it to a repository. ## Set up your client Run this in a terminal. ```bash Terminal theme={null} claude mcp add --transport http signal https://first-signal-api.fly.dev/mcp \ --header "Authorization: Bearer sig_your_key" ``` By default Claude Code keeps the server for you, in the current project only. Add `--scope user` to use it in every project. Avoid `--scope project`: it writes the server, key included, to a `.mcp.json` file that you would share through git. Check the connection with `claude mcp list`, or run `/mcp` inside Claude Code. Then ask: > Use Signal's whoami tool and tell me what this key can do. Claude Desktop starts MCP servers as local programs from its config file, so it reaches the hosted endpoint through `mcp-remote`, a widely used npm package that speaks stdio to Claude Desktop and Streamable HTTP to Signal. Signal does not make or support it. It needs Node.js. Open **Settings**, then **Developer**, then **Edit Config**. That opens `claude_desktop_config.json`: * macOS: `~/Library/Application Support/Claude/claude_desktop_config.json` * Windows: `%APPDATA%\Claude\claude_desktop_config.json` Add the server, keeping any servers you already have. ```json claude_desktop_config.json theme={null} { "mcpServers": { "signal": { "command": "npx", "args": [ "-y", "mcp-remote", "https://first-signal-api.fly.dev/mcp", "--header", "Authorization:${SIGNAL_AUTH}" ], "env": { "SIGNAL_AUTH": "Bearer sig_your_key" } } } } ``` The header value sits in `env` so the argument has no space in it, which some systems mishandle. This is the same snippet the **Claude Desktop** tab in the app shows, with your key filled in. Quit Claude Desktop completely and open it again. Cursor reads `~/.cursor/mcp.json` for every project, or `.cursor/mcp.json` in one project. ```json .cursor/mcp.json theme={null} { "mcpServers": { "signal": { "url": "https://first-signal-api.fly.dev/mcp", "headers": { "Authorization": "Bearer sig_your_key" } } } } ``` In a project file that you commit, keep the key out of the file. Cursor fills in environment variables. ```json .cursor/mcp.json theme={null} { "mcpServers": { "signal": { "url": "https://first-signal-api.fly.dev/mcp", "headers": { "Authorization": "Bearer ${env:SIGNAL_API_KEY}" } } } } ``` Open the MCP section of Cursor's settings and check that `signal` is on and lists its tools. Then ask the agent to call `whoami`. VS Code reads `.vscode/mcp.json` in a workspace. Its file uses `servers`, not `mcpServers`. The `inputs` block makes VS Code ask for the key the first time the server starts and keep it out of the file, which you will probably commit. ```json .vscode/mcp.json theme={null} { "inputs": [ { "type": "promptString", "id": "signal-api-key", "description": "Signal API key", "password": true } ], "servers": { "signal": { "type": "http", "url": "https://first-signal-api.fly.dev/mcp", "headers": { "Authorization": "Bearer ${input:signal-api-key}" } } } } ``` Select **Start** above the server in the file, or run **MCP: List Servers** from the command palette. Enter your key when asked. The tools are available in Chat in agent mode. Any client that supports MCP servers can connect. Which of these it takes depends on the client. **A client that speaks Streamable HTTP and can send a header** connects to the hosted endpoint. Most use a JSON file shaped like this. The key names vary a little, so check what your client calls the URL and headers. ```json mcp.json theme={null} { "mcpServers": { "signal": { "type": "http", "url": "https://first-signal-api.fly.dev/mcp", "headers": { "Authorization": "Bearer sig_your_key" } } } } ``` **Windsurf** reads `~/.codeium/windsurf/mcp_config.json`. For a remote server it may expect `serverUrl` where the file above has `url`. Signal has not tested it in Windsurf, so check Windsurf's current documentation if it does not connect. **A client that only starts local programs** can use the same `mcp-remote` bridge as Claude Desktop above. Give it the command `npx`, the arguments `-y mcp-remote https://first-signal-api.fly.dev/mcp --header Authorization:${SIGNAL_AUTH}`, and `SIGNAL_AUTH` set to `Bearer sig_your_key` in its environment. **A client that supports only the older SSE transport** cannot use the hosted endpoint. The endpoint answers `GET` with `405`, because it keeps no session and opens no stream. To test the endpoint without a client, ask it for its tool list. ```bash Terminal theme={null} curl -sS https://first-signal-api.fly.dev/mcp \ -H "Authorization: Bearer $SIGNAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' ``` ## If it does not work | What you see | Cause and fix | | - | - | | `401` with `mcp_unauthorized`, or "unauthorized" in the client | The key is missing, wrong or revoked. The header must read `Bearer ` followed by the key, with one space. Check the key's row on the **API keys** page | | The write tools are not in the list | You are using a read key on the hosted endpoint, which hides them. Make a **Write** key if you want them | | A tool answers `agent_required` | The key is for all agents. Pass `agentId` (ask the assistant to call `list_agents` first) or make a key tied to one agent | | A tool answers `403` with `key_read_only` | A read key called a write tool. The hosted endpoint hides the write tools from a read key, so this is rare. Make a **Write** key if you need them | | A tool answers `402` with `quota_exceeded` | Your plan's simulated test calls for the period are used up, so a run was refused whole. Run fewer tests, upgrade, or wait for the reset. See [Limits and quotas](/limits#plan-limit-errors) | | A tool answers `429` | You reached a rate limit. Wait the number of seconds in `Retry-After`. See [Rate limits](/limits#api-rate-limits) | | Claude Desktop shows the server as failed | Node.js is not installed or not on the path, or `SIGNAL_AUTH` does not read `Bearer ` followed by the key. Open the MCP log from **Settings**, then **Developer** | | The client cannot connect, or reports a `405` | The client expects the older SSE transport. Use a client or bridge that speaks Streamable HTTP | If you report a problem, include the `requestId` from the tool error. ## Next steps Every tool, example requests, and what a read or write key can do. The same operations over plain HTTP. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.