Guard Core MCP¶
guard-core-mcp is an MCP server that lets AI coding agents answer questions about the Guard security ecosystem (fastapi-guard, guard-core and guard-agent) by introspecting the libraries actually installed in your project, instead of answering from training data.
Why¶
An agent can already read the docs. What it cannot do from memory is tell you that redis_failopen in your config is silently doing nothing because the real field is redis_fail_open, that a flag did not exist until guard-core 3.5.0, or whether a given request would actually be blocked and by which pattern. Pydantic ignores unknown keys instead of raising, so a misspelled setting fails silently at runtime with no error anywhere.
guard-core-mcp answers those questions from the packages installed in the interpreter running it: real Pydantic model validation, real field metadata pulled from the model itself, and requests run through guard-core's actual detection engine. When a library is not installed, the affected tools say so explicitly instead of guessing, see Installation for what that looks like.
The Six Tools¶
| Tool | Answers |
|---|---|
versions |
Which Guard libraries are installed here, and at what version, alongside the versions the bundled documentation covers |
validate_config |
Is this config dict valid against the real SecurityConfig/AgentConfig model, including typo'd keys Pydantic would otherwise ignore, and deprecated fields |
config_fields |
What does this setting do, what's its type and default, or does a setting for X exist at all |
search_docs |
Which bundled documentation page covers this, with a citable live URL |
get_doc |
The full text of one bundled documentation page |
check_payload |
Would this request be blocked by guard-core's detection engine, and by which pattern |
See Tools for the signature, parameters, and an example call and response for each.