How does context7 decide which docs are trusted? What is trustScore?
I'm using the context7 MCP server (upstash/context7). A GitHub issue (April 2025) reported that unrestricted resource uploads pose a security risk and create duplicate entries. How did the maintainers respond, and what does the trustScore they introduced actually control — does it affect which docs are served via MCP?
Originally asked on GitHub: https://github.com/upstash/context7/issues/29
5 answers
Status: explained — the issue is closed as completed, and the maintainers introduced
trustScoreto address the permissive upload policy.The original issue (opened 2025-04-20) reported that anyone could upload any resource type, which was considered a security risk and produced duplicate entries; it asked for stricter validation.
A contributor agreed that defining "appropriate" resources is hard without human intervention and proposed an
approvedflag: approved docs would appear in MCP, the API, and the web; unapproved docs would appear only on the web. The initial criterion was more than 30 GitHub stars. Another commenter objected that stars can be farmed and aren't a reliable reputation metric; the contributor agreed and asked for suggestions. A later suggestion was to run heuristics on the underlying artifacts — stars as one factor among several, account reputation, repo activity — plus controls like pinning repos to specific versions/hashes and blocking typosquatting.On 2025-05-23, a contributor stated that context7 introduced
trustScore, based on multiple criteria such as stars and activity, and that user-defined policies for MCP filtering would be enabled. A follow-up comment (2025-06-24) asked where trustScore can be seen and whether it ranks MCP server results; the thread contains no answer.The thread does not confirm whether the originally proposed
approvedflag details (visibility rules for MCP/API/web, the 30-star threshold) were implemented as described, nor how trustScore relates to the MCP results you see.Security note: this issue was filed because permissive resource uploads were considered a security risk that could be exploited for malicious purposes.
Did this work for you?What @enesakar tried on GitHub (Outcome unclear)
A contributor announced that context7 introduced trustScore, based on multiple criteria (stars, activity, etc.), and that user-defined policies for MCP filtering would be enabled later.
Source: https://github.com/upstash/context7/issues/29#issuecomment-2906048779 (opens external site)
Did this work for you?What @eladcandroid tried on GitHub (Outcome unclear)
The issue reporter described the current behavior as unrestricted uploading of any resource type, posing security risks and creating duplicate entries, and requested stricter validation rules.
Source: https://github.com/upstash/context7/issues/29 (opens external site)
Did this work for you?What @lirantal tried on GitHub (Outcome unclear)
A commenter suggested running heuristics on the underlying artifacts: stars as one of several factors, account reputation, repo activity, pinning repos to specific versions/hashes, and blocking typosquatting.
Source: https://github.com/upstash/context7/issues/29#issuecomment-2817233475 (opens external site)
Did this work for you?What @enesakar tried on GitHub (Outcome unclear)
A contributor proposed an approved flag: docs with more than 30 GitHub stars would be listed in MCP, APIs, and web; unapproved docs would appear only on the web.
Source: https://github.com/upstash/context7/issues/29#issuecomment-2817199908 (opens external site)
Did this work for you?
Related questions
- Cursor MCP "Failed to create client" with @upstash/context7-mcp using npx: fix
- Context7 MCP server rate limits my first request in Trae — is it my config or Context7?
- Does Context7 support GitLab repositories now, and where do I add one?
- Is there an official public Docker image for the context7 MCP server?
- Context7 MCP in Cursor: "Client closed" / "spawn npx ENOENT"