How Remote OpenClaw Scores MCP Servers
Every MCP server in our directory gets a quality score from 0 to 100 and a letter grade. The score is computed deterministically from data we already collect — GitHub stars, update recency, published metadata, and provenance. We never fabricate signals or make external calls at score time, so the same server always produces the same score. Here is exactly how it works.
The five dimensions
Scores are the sum of five weighted dimensions. Popularity and maintenance measure momentum; completeness and trust measure how safe and ready a server is to adopt.
25
Popularity
20
Maintenance
20
Completeness
15
Trust
20
Security
Popularity
Max 25 ptsDerived from GitHub stars, a durable proxy for real-world adoption and community trust.
| 10,000+ stars | 25 |
| 1,000–9,999 stars | 20 |
| 100–999 stars | 13 |
| 10–99 stars | 7 |
| 1–9 stars | 3 |
| 0 or unknown | 2 |
Maintenance
Max 20 ptsHow recently the source repository was updated, measured against our fixed reference date of 9 July 2026.
| Updated < 90 days ago | 20 |
| Updated < 6 months ago | 14 |
| Updated < 1 year ago | 8 |
| Over a year, or stale | 3 |
| Unknown | 5 |
Completeness
Max 20 ptsHow thoroughly the server documents itself. Points accumulate across five metadata checks.
| One-line install command | +7 |
| Declared license | +5 |
| Description of 80+ characters | +4 |
| Env vars or known transport | +2 |
| Category or tags present | +2 |
Trust
Max 15 ptsProvenance and verifiability — whether the server is official, registry-listed, or a public community project.
| Official server | 15 |
| Active in the MCP registry | 9 |
| Public source repository | 6 |
| Limited provenance | 3 |
Security
Max 20 ptsDeterministic deductions from a starting 20, based on the server's declared credential surface and its known tool schema (from README docs or a live MCP handshake). It is a signal, not a code audit.
| Starting score | 20 |
| No declared license | −3 |
| stdio transport handling 3+ env credentials | −3 |
| 5+ env credentials | −2 more |
| 3+ destructive-named tools (delete, exec, rm…) | −4 |
| Tool surface not yet analyzed | capped at 12 |
| Behavior verified via live MCP handshake | +2 (max 20) |
Grade bands
The five dimensions add up to a total out of 100, which maps to a letter grade:
85–100
Excellent — well adopted, actively maintained, fully documented, and verifiable.
70–84
Strong — solid on most dimensions with minor gaps.
55–69
Fair — usable, but with notable gaps in adoption, docs, or upkeep.
Below 55
Weak signals — evaluate carefully before relying on it.
Show off your score
Every scored server has an embeddable badge you can paste into your GitHub README. It renders live from our domain and links back to the server’s page. Grab the exact snippet from the Score section of any MCP server page. The Markdown looks like this:
[](https://www.remoteopenclaw.com/mcp/{owner}/{repo})
Live exampleWhat the score is not
The score reflects public, measurable signals — not a code audit or a security review. A high score means a server is popular, maintained, well-documented, and verifiable; it does not certify that the server is bug-free or safe for your specific credentials. Always review the source before granting an MCP server access to sensitive data.
Browse scored MCP servers →