Weckr

@weckr/mcp

Query your Weckr AI cost and margin data directly from Claude, Cursor, or any other MCP-compatible AI assistant.

Try it live

See what Weckr looks like before connecting your own data. 👉 https://app.useweckr.com/demo

What you can ask

Ask things like:

• *"Which of my users are losing me money this month?"*
• *"What's my AI margin right now?"*
• *"Which feature should I switch to a cheaper model?"*
• *"Is my pricing sustainable given my actual AI costs?"*
• *"Show me the feature cost breakdown."*

Claude calls the Weckr API and gives you the real numbers from your dashboard, in the conversation.

Get an api key

Sign up at app.useweckr.com. Your wk_ key is shown once when you create a project. Copy it. If you missed it, mint a new project at app.useweckr.com/dashboard/projects/new.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{

"mcpServers": { "weckr": { "command": "npx", "args": ["-y", "@weckr/mcp"], "env": { "WECKR_API_KEY": "wk_your_key_here" } } } }

Restart Claude Desktop. Try: *"Use the Weckr tools to show me my AI costs this month."*

Why `npx` and not `npm install -g`? Claude Desktop runs under launchd on

macOS, which doesn't inherit your interactive shell's PATH. If you installed

Node via nvm / asdf / volta, a globally-installed weckr-mcp binary

usually isn't on Claude Desktop's PATH and the server fails to start with no

diagnostic. npx -y @weckr/mcp sidesteps the PATH problem entirely.

Cursor

Add to .cursor/mcp.json at your project root (or ~/.cursor/mcp.json for global):

{

"mcpServers": { "weckr": { "command": "npx", "args": ["-y", "@weckr/mcp"], "env": { "WECKR_API_KEY": "wk_your_key_here" } } } }

Tools exposed

Configuration

How it works

The MCP server is a tiny stdio process spawned by your AI client. It calls the Weckr HTTP API using your wk_ key (server-to-server, no JWT needed). The api key authoritatively identifies one project. The server uses it to resolve projectId once at startup, then scopes every subsequent call to that project. The dashboard endpoints (/api/v1/stats, /users, /recommendations/*) verify the URL project id matches the api key's project and return 404 on mismatch, so a leaked api key cannot be used to read another project's data.

No prompt text or completion text is sent anywhere by this package. The tools only read aggregated stats. The api key never leaves your local process *except* to the URL configured via WECKR_BASE_URL (default app.useweckr.com). If you change WECKR_BASE_URL, the SDK prints a stderr warning so you can spot a copy-paste config that points at the wrong host.

Troubleshooting

Claude Desktop shows the weckr server as disconnected, with no error. Likely a PATH issue. Switch to the npx -y @weckr/mcp form above. To see the actual error from the spawn, run the server manually in a terminal: WECKR_API_KEY=wk_... npx -y @weckr/mcp. Anything wrong (missing api key, 401 from Weckr, network issue) prints to stderr.

`Weckr API 401 on /api/v1/me`. Your WECKR_API_KEY is empty, malformed, revoked, or pasted with a trailing space. Confirm a working key by visiting app.useweckr.com/dashboard/projects/new and creating a fresh project.

Tools show "Project not found or forbidden" (404). You set WECKR_PROJECT_ID to a UUID that doesn't belong to the project the api key owns. Remove WECKR_PROJECT_ID so the server resolves the right project id from the api key.

Local development

git clone https://github.com/Ghiles3232/weckr-sdks

cd weckr-sdks/mcp npm install npm run build

Smoke-test against production:

WECKR_API_KEY=wk_... node dist/index.js

(prints "weckr-mcp running on stdio" to stderr, then waits for JSON-RPC on stdin)

To point at a local Weckr backend:

WECKR_API_KEY=wk_... WECKR_BASE_URL=http://localhost:3000 node dist/index.js

Source

Open source at github.com/Ghiles3232/weckr-sdks.

License

MIT