Two claims make it worth adding: no auth at all, and an implementation the author puts at around 280 lines with 18 test cases — so the thing answering your cost questions is auditable in one sitting. The pricing table is hand-checked, which is its strength and its limit: verify anything you are about to base a budget on against the provider's own page.
A hosted MCP server over a hand-checked LLM pricing table. It answers cost questions directly: what a model charges, what a given prompt would cost, and which model is cheapest for a described job.
- The model list and one model's pricing record — `list_models`, `get_model`
- Two models compared on price — `compare_models`
- The cost of a specific prompt or workload — `estimate_cost`
- The cheapest model for a stated need, and the providers covered — `cheapest_for`, `list_providers`
- The raw table as resources — `usagewall://pricing/full` and `usagewall://pricing/<provider>`
- A public JSON pricing API and a status page alongside the MCP endpoint, for use outside a client
Nothing installed and nothing to authorize: the server is at `https://usagewall.vercel.app/api/mcp`, read-only and free, spoken over JSON-RPC 2.0 on streamable HTTP. In Claude.ai it is added as a custom connector by pasting that URL; install notes for Claude Desktop, Cursor, Windsurf, Zed and Cline are on the project's `/mcp` page. You can check it before adding it by posting a `tools/list` request to the endpoint with curl.