← All rules

mcp-scan rule

Known API endpoint

network-egress-apiinfo

What mcp-scan saw

An endpoint's host matched the api category in KNOWN_ENDPOINTS by exact host or dot-suffix: api.openai.com, api.anthropic.com, api.github.com, api.hubapi.com, api.mistral.ai, generativelanguage.googleapis.com, api.groq.com, api.replicate.com, api.together.xyz, api.cohere.ai, api.perplexity.ai, api.deepseek.com, api.fireworks.ai, api.voyageai.com, api.stability.ai, or api.elevenlabs.io.

Why it fires

It is inventory, graded INFO with no fixRecommendation in the source. Its real function is to keep recognized model and platform APIs out of the network-egress-unknown bucket, so that the unknown list stays short enough to read. Registered-domain matching means api.openai.com.evil.io is not classified here.

When this is a false positive

  • Nothing to dismiss. This rule reports that a destination is recognized, not that anything is wrong.
  • It does say something worth checking: a server contacting a model API you did not expect it to use is worth a look, and this finding is where you would notice.
  • The list is hardcoded with a fixed set of hosts, so an API you consider equally standard reports under network-egress-unknown instead. That difference is about the list, not about the host.

How to fix it

  1. Confirm the API matches what the server is for. A filesystem server contacting a model API is worth explaining.
  2. Check where the credential for that API comes from. If it is a literal in the config, the exposed-secret finding will be alongside this one.
  3. Add the host to allowedDomains in .mcp-scan.json if you want it excluded from the AST scanner's external-endpoint checks as well.
  4. No action is required for the finding itself. It carries no fix recommendation because there is nothing to fix.