safe-mcp sign in

library ยท context7

Context7: two read tools, no sign-in, logged like everything else.

Context7 returns current documentation for a library, so a coding agent stops guessing at an API from training data. It has two tools, both reads, and needs no account. The gateway adds exactly one thing here: the record.

the entry
nameio.github.upstash/context7
endpointmcp.context7.com/mcp
authorizenone; an API key raises Context7's rate limit
toolsresolve-library-id, query-docs
writesnone

Defaults as shipped. Tool names from Context7's README.

What changes when it runs through the gateway.

Every query and every answer is in your project's log with the reason the agent gave, so a month later you can see which documentation a change was based on. Reads always run, with or without a goal, so Context7 works in a project that has no goal at all.

What leaves the gateway: the library name and the query, to Context7. Nothing else.

One caveat.

Whether the two tools count as read-only to the gateway is decided by Context7's own annotations; a library reference cannot add names. If they are not marked, they still run as long as a goal stands, and they are blocked in testing mode. A custom entry with the same URL and both names listed as read-only removes that wrinkle.

Bring the rest yourself.

Any MCP server reachable over HTTP is one custom entry in the project file, with its read-only tools named by you.