The reliable MCP registry
MCPfinder is a trusted directory of Model Context Protocol servers. We’re not the biggest. We’re the one your CTO won’t reject.
Our mission
Model Context Protocol is how modern AI agents reach the rest of your stack — your database, your repo, your ticketing system, your docs. A bad MCP server is a production incident with a chat interface. MCPfinder exists so that engineering teams can find MCP servers that have been vetted for license, repo health, documentation, release cadence, and recent activity — and skip the ones that haven’t.
Readers never pay. The registry is funded by vendors who buy placements, verified badges, and editorial — transparently labeled, never hidden. See how advertising works.
What makes us different
Repo reachable, permissive license, README present, at least one release or tag, parseable install command, recent activity within 180 days. No exceptions.
Install commands per package manager, usage examples, version history, license, and last-checked timestamp. Not one-sentence cards.
Filter by language, runtime, license, auth model, and last-release window. Instant results. Powered by Meilisearch.
Side-by-side compare pages and 'alternatives to' lookups for every listing. We help you pick, not just scroll.
Vendors fund the registry. Every placement, backlink, and sponsored article is explicitly labeled — in the UI and in the code.
A scheduled reliability check hides MCPs that break, lose their license, or go 180 days without a commit. Authors get notified first.
Who we serve
- SaaS founders and CTOs evaluating MCP servers to embed in their products or give to their engineering teams.
- Engineering teams at mid-size companies who need to know an MCP is maintained before they wire it into an internal agent.
- AI-native startups building on Claude, Cursor, Windsurf, and Zed — where picking the wrong MCP shows up in customer support tickets an hour later.
- MCP server authors who want their work found by buyers, not just listed on a wall.
Editorial principles
- Paid never outranks broken. Placements buy visibility — not reliability. A flagged listing is hidden from search regardless of tier.
- Sponsored content is labeled. Always. In the card, in the URL metadata, and in the page body.
- No AI-written listings. Descriptions, usage examples, and reviews are written by humans or pulled from canonical sources (the repo’s own README, the package manifest).
- Backlinks are auditable. Every outbound link has a row in our database with anchor, placement, and follow status. If you’re wondering how we treat SEO, we treat it the way Google wants us to.
Free submission if it passes the reliability gates. Paid tiers for featured placement, verified badge, and editorial.