Weckr

What the weckr MCP server does

The weckr MCP server gives supported AI clients access to Weckr’s cost and margin analysis for LLM-powered SaaS products. Its primary use is answering questions about spend by customer, feature, or model and identifying users or plans where AI costs exceed revenue.

Weckr is designed for applications that need to understand the economics of individual AI requests. The broader repository includes SDKs for TypeScript/Node and Python, while the MCP distribution is published as @weckr/mcp. The README identifies Claude and Cursor as supported MCP clients.

The service is intended for SaaS founders and developers assessing whether AI features are profitable. It can also support model-cost reviews by surfacing cheaper model recommendations in the Weckr dashboard.

How it works

The SDK-based flow wraps an LLM call with wk.chat(client, opts). The wrapped request is sent to OpenAI, Anthropic, or Gemini and returns the original result. After the call completes, the SDK sends metadata to the Weckr API without adding work to the request path.

Recorded metadata includes a user identifier, feature, model, token counts, latency, and plan. Weckr calculates cost on its backend using public per-token pricing, then stores cost, revenue, and margin for each request. Results are rolled up by user, feature, and model for dashboard analysis.

The service does not receive prompt or completion text. Its documented telemetry consists of call metadata and the supplied userId string. This makes the MCP server a fit for cost analysis rather than prompt inspection or application-content retrieval.

Setup and configuration

Install the MCP package with:

npx -y @weckr/mcp

The repository lists the MCP package under its Claude/Cursor distribution. It also provides a separate Claude Code plugin installed with /plugin install weckr@weckr, but that plugin is a skills package rather than the MCP package itself.

An API key is available from app.useweckr.com, which offers a free tier without requiring a credit card. The README does not specify MCP environment variable names or client configuration fields, so those settings should be taken from the package’s detailed setup documentation.

Tools and capabilities

The documented capabilities include:

• Querying LLM cost by customer, feature, and model.
• Finding users whose LLM expense is higher than their revenue.
• Reviewing cost, revenue, and margin per request.
• Examining model pricing and cheaper-model recommendations through Weckr’s analysis.
• Tracking metadata from calls made through the TypeScript or Python SDKs.

The README does not provide a formal MCP tool-name list, so tool names and exact argument schemas should be verified in the package documentation before writing client-side calls.

Limitations and notes

The MCP README excerpt does not document the individual MCP tool schemas, authentication parameters, transport settings, or required environment variables. It also does not claim that the MCP package itself performs SDK instrumentation; the described metadata collection is implemented by the TypeScript and Python SDKs.

Weckr’s backend computes costs from public per-token pricing. Pricing data is maintained through a live feed and a weekly review process, but model availability and rates can change. The system records metadata rather than prompts or completions, and the userId value is supplied by the integrating application.

The weckr MCP server is most relevant when an application already uses Weckr data or needs an AI-assisted view of LLM unit economics. It is not documented as a general observability, tracing, prompt-management, or billing provider.