Model picker for agents (MCP)

The same comparison you run in the model picker is available to AI agents as a Model Context Protocol tool.

• Address: https://measuredlab.com/api/model-picker/mcp (Streamable HTTP; protocol versions 2025-11-25, 2025-06-18 and 2025-03-26)
• Tool: measuredlab_pick_models, read-only, no account and no key needed
• Limits: 30 requests per minute and 1,000 per day from one address; over the limit the server answers 429 with Retry-After

Send numbers and requirements only

The tool takes structured parameters: request volume, token sizes, deadline, required features, data rules. Never send task descriptions, prompts, documents, customer data or API keys. Volumes can reveal business information, so tell your user before the call.

Ask the user to confirm first

Show the parameters to the person and set parameters_confirmed: true only after they agree. Without confirmation the tool returns an error and calculates nothing.

Pin the data version

Every answer carries dataset_sha256. Pass it back to repeat a comparison on the same data; if that version is no longer served, the tool says so instead of silently using newer prices.

What the answer is

Prices and limits from dated sources, each with its link and check date, and the cost arithmetic for your parameters. It is not a guarantee of task quality: test the shortlisted models on your own examples.

Client configuration

{ "mcpServers": { "measuredlab": { "url": "https://measuredlab.com/api/model-picker/mcp" } } }