Why we recommend these tools
We match tools from our curated list to what you need to do.
Consumer recommendations
We map your request to a task or tool type and only consider tools in matching categories. On the website, a relevance model then scores those candidates. Published tasks use curated picks; the public API uses deterministic matching. When no specialist fits, we say so and point to general assistants. Popularity is an editorial 1–100 score we set for each tool from its public reach when it enters the catalog — not live usage data. It adds 15% to the relevance score and orders browse lists. We have not hands-on tested these tools yet.
Affiliate relationships and sponsorship never change organic rank. Paid placements are labeled separately. About Agent Rerank
Developer: MCP server ranking
Registry listings, repository freshness and stars, and task relevance are available today. Connection and security testing is not yet published for most servers. The formula below describes how recorded signals are combined; missing evidence is not a passing test.
Ranking signals
| Signal | Weight | What it measures |
|---|---|---|
| Relevance | 65% | Retrieval plus a dedicated relevance reranker scores the task match. |
| Health | 12% | The latest recorded check for the server. |
| Freshness | 8% | Repository activity, decayed over 90 days. |
| Security | 8% | Grades A through F map to a normalized signal. |
| Token footprint | 4% | Smaller approximate tool footprints score higher. |
| GitHub popularity | 3% | Stars, logarithmically scaled. |
Inspect the evidence
Scorecards show MCP handshake and tools/list results, security grade and reasons, approximate token footprint, health, freshness and measurement dates. Missing or malformed evidence is unknown, never a passing test. Handshake and tools/list results provide protocol evidence on the scorecard; they are not separate weighted ranking signals.
The live ranking path uses retrieval, a dedicated relevance reranker and deterministic signal blending. It does not use an LLM judge. Scores are comparative signals, not a guarantee of safety or suitability.
Pages with fewer than three results carrying measured ranking signals are marked noindex. Sponsored placements never alter organic rank.
The consumer side works differently: tools come only from our curated catalog, matched to your words by relevance and popularity. Read how that side works.
Votes reorder community lists; they never change the #1 pick for your task.