How we keep rankings honest.
Last updated: June 15, 2026
1. Selection criteria
We add servers to MCP Select based on three things: community interest (GitHub stars, npm downloads, registry presence), relevance to a common workflow (browser automation, databases, search, etc.), and whether we can actually run it. We do not list servers we cannot install or that lack a public repository.
2. Independence
Testing is fully independent. No vendor can pay for a better score, a higher rank, or a positive review. We buy our own accounts when needed and run every server from a clean environment with no cached credentials.
3. What we publish
We publish pass rates, latency, install configs, and failure reasons. If a server fails every task, we show that. If it requires an API key we do not have, we label it "Researched" and explain what we could not test.
4. Update cadence
We rerun benchmarks at least every two weeks and after any major MCP SDK release. Each result page shows the test date. Stale results are labeled, not removed.
5. Corrections
If we misrepresent a result, fix a bug in our harness, or receive evidence that a score is wrong, we update the page and note the correction in our changelog.
6. Funding and affiliates
MCP Select is operated by Pixel Investment UG, Berlin. We do not currently run active affiliate programs. If we add affiliate links or sponsored listings in the future, they will be clearly labeled and will never influence test scores or organic rankings. See our Affiliate Disclosure.
7. Reproducibility
All benchmark scripts and raw data are open source at github.com/AlexGn/mcpselect. Anyone can rerun the tests and verify the results.