MCP100 is an independent, evidence-led guide to Model Context Protocol (MCP) servers and related tools. We help people discover relevant options, understand the evidence behind them, and choose tools that fit real tasks.
The question is not simply “Which MCP is the most popular?” It is:
Which MCP is the right choice for this job—and what evidence supports that choice?
Being visible in a directory is not the same as being recommended. MCP100 makes that difference explicit.
Why MCP100 exists
New MCP servers appear regularly. Existing projects change publishers, add capabilities, alter permissions, become inactive, or disappear. A large directory can help you find options, but it cannot by itself tell you whether a particular tool is appropriate for your work.
Before choosing an MCP, people need practical answers:
- What task does it actually help accomplish?
- Can its implementation and publisher be identified?
- What permissions or sensitive access does it require?
- Is it maintained, and what are its operating limitations?
- What claims are supported by documentation or other attributable evidence?
- What is still unknown?
MCP100 uses editorial review to make those distinctions clearer.
Our three editorial levels
MCP100 uses three public levels. Each represents a different editorial conclusion and standard of evidence—not a popularity ranking.
Recommended — Evaluated for a defined use case
A Recommended MCP has met the applicable recommendation requirements for a defined task at the time of review.
Evaluation considers implementation identity, publisher relationship, MCP relevance, task fit, documentation, limitations, maintenance, tool behavior, permissions, installation and supply-chain concerns, security and operational evidence, and differentiated value. The relevant requirements must be supported by attributable evidence rather than popularity alone.
What this means: MCP100 has sufficient evidence to recommend this MCP for a specific purpose. It does not guarantee security, future maintenance, compatibility, or suitability for every environment.
Shortlist — Promising, with documented evidence gaps
A Shortlist MCP has a credible implementation identity and meaningful task value, but has not yet met every applicable requirement for recommendation.
Shortlist candidates must not have disqualifying material admission failures or security vetoes. Their outstanding recommendation gaps are identified, and a future review is scheduled.
What this means: This MCP is worth deeper consideration, but it is not yet Recommended. Shortlist status is not approval for production use.
Watchlist — Worth tracking, not yet recommended
A Watchlist MCP has enough reviewed, attributable evidence to justify continued public tracking.
This requires a resolved implementation identity, meaningful MCP relevance, an evidence-backed signal worth following, an assessment of material blockers to public tracking, and a specific reason and date for future review. Marketplace rank or popularity alone is not sufficient.
What this means: This MCP is relevant enough to monitor and revisit. It is not an installation recommendation, a security clearance, or an assurance of production readiness.
Being tracked is not the same as being shortlisted. Being shortlisted is not the same as being recommended.
How editorial decisions change
Classifications reflect the evidence available at the time of review. A project may qualify for a different level as stronger evidence emerges, or when its identity, permissions, ownership, maintenance, or risk profile changes.
- A Watchlist MCP can qualify for Shortlist or Recommended if it meets the applicable review requirements.
- A Shortlist MCP can become Recommended only after the applicable recommendation requirements have been met.
- A Recommended MCP may be reassessed and placed at a lower level when the evidence no longer supports its prior classification.
- A project may leave the public directory if it does not meet the minimum standard for continued inclusion.
These are editorial decisions based on review, not automatic rewards for gaining stars or directory rankings. A serious security or identity concern may prevent inclusion at any level. An MCP does not have to pass through every level in a fixed sequence.
What we evaluate
Depending on the level and type of resource, our reviews consider:
- Practical task fit and differentiated value
- Implementation identity and publisher provenance
- Documentation, capabilities, and known limitations
- Requested permissions and tool behavior
- Maintenance and operational health
- Installation and software supply-chain considerations
- Relevant security evidence and its freshness
- Supported clients and integration requirements
- Attributable adoption or activity signals
Not every dimension has equal weight at every level. Unknown is not the same as verified, and a claim from a publisher is not the same as an independently verified fact.
Where useful and feasible, MCP100 may perform limited, controlled technical checks of specific claims. Such checks are not comprehensive security audits or certifications.
Our editorial principles
Evidence before scores
Attributable evidence matters more than popularity metrics. Rankings help organize results; they do not determine recommendation status.
Fit before popularity
The most popular MCP is not necessarily the best fit for a particular task.
Unknown means unknown
Missing evidence is not treated as proof that a tool is safe, compatible, or maintained.
Decisions can change
We reassess editorial conclusions as evidence and projects evolve.
Commercial independence
Payment cannot buy a recommendation.
Advertising, sponsorships, partnerships, affiliate relationships, and commercial agreements do not determine whether an MCP is Recommended, Shortlist, or Watchlist. Sponsored placements, if any, are clearly identified and kept separate from editorial classifications.
MCP100 is an independent project and is not affiliated with Anthropic or MCP publishers.
Transparency
Editorial decisions should be understandable. Where possible, MCP100 shows supporting sources, review dates, limitations, open questions, and setup guidance so people can see why an MCP has its current classification—not simply that it appears in the directory.
Frequently asked questions
Why aren't there exactly 100 Recommended MCPs?
MCP100 is evidence-based, not quota-driven. The number of Recommended MCPs depends on how many meet the editorial requirements, up to the collection's intended limit. We will not recommend a tool merely to fill a slot.
What's the difference between Shortlist and Watchlist?
Shortlist candidates have met stricter admission requirements and have specific, documented gaps remaining before recommendation. Watchlist candidates have met the separate, lower standard for public tracking, but may need more evidence before deeper consideration. Neither level is a recommendation.
Does every MCP have to move from Watchlist to Shortlist to Recommended?
No. Levels reflect current evidence and editorial decisions, not required steps in a sequence. A candidate may enter the level its evidence supports after the appropriate review.
Can companies pay to be Recommended?
No. Commercial relationships do not determine editorial classifications.
Why isn't a particular MCP included?
Some projects have not been discovered or reviewed. Others may lack reliable identity or meaningful evidence, or may not meet the minimum requirements for public tracking. Not appearing on MCP100 is not, by itself, a judgment that a project is unsafe or unsuitable.
How often are classifications reviewed?
Review timing depends on the evidence, the level, and relevant changes to the project. Review dates and updated evidence may be shown on individual MCP pages when available. No classification is a permanent guarantee.
Our approach
Discover broadly. Evaluate carefully. Recommend selectively.
MCP100 aims to make a broad ecosystem easier to explore while keeping its recommendations selective, explainable, and grounded in evidence.