What it does
ModelShortlist runs locally as an MCP server and gives an AI assistant evidence for choosing a model. The evidence covers the full OpenRouter model catalog, including tool and function-calling support, context and completion limits, and input and output pricing, together with Artificial Analysis Intelligence, Coding and Agentic indexes and pricing and performance figures. Zero Data Retention (ZDR) is applied as a hard constraint only when it is explicitly required. Upstream data is refreshed on demand or when the cache expires. The server has no hosted backend and no account, and it sends no telemetry from the MCP. API keys stay with the local process, which calls the upstream services directly. The host AI makes the final recommendation; the tool does not apply a fixed ranking formula.
It is intended for choosing a model against a specific workload, cost, context, tool-use or privacy requirement. The recommend_models tool filters by hard constraints such as ZDR, tool calling, minimum context and maximum price. compare_models returns catalog, ZDR and benchmark data for a chosen set of OpenRouter model IDs. modelshortlist_status reports source freshness, catalog and ZDR coverage, match coverage and cache state. Each source is marked fresh, stale or unavailable, so outdated data is not presented as current. If Artificial Analysis fails, OpenRouter models remain listed with benchmark fields missing rather than scored as zero.
Setup uses npm through npx, or a guided local setup that stores both API keys in a gitignored .env.local file and prints MCP configuration for the client. A browser-only configurator generates configuration for Claude Desktop, Hermes Desktop, Cursor, Claude Code and VS Code / Copilot, and keys entered there are not sent to a server. The package is published under the MIT license.
Limitations and requirements
Requires an Artificial Analysis API key and an OpenRouter API key, Node.js 20 or later, and an MCP-capable client. On Windows GUI clients where npx is not on the application PATH, npx.cmd or the guided setup is needed. Benchmark data is attached only when a model can be confidently matched; unmatched models stay listed with missing benchmark fields. The publisher notes that OpenRouter ZDR metadata can be unavailable for ordinary requests.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 17 Nov 2026 · its page on the source
ModelShortlist
Local BYOK MCP for AI model shortlisting using OpenRouter + Artificial Analysis.
v0.2.5As shown on registry.modelcontextprotocol.io when we last read it