Comparison¶
passmcp alone answers "is this server safe"; a general graph database answers anything you load into it. passmcp-graph answers the questions in between — which agents reach what, under which identities — from passmcp's own evidence, with no service to run.
| Approach | Reads passmcp's evidence directly | Needs a running service | Policy gate with an exit status | Holds agents and identities | Network access |
|---|---|---|---|---|---|
| passmcp-graph | yes | no | yes: check --policy, exit 1 on a violation |
yes | none |
passmcp check, one server at a time |
yes | no | per server | no agents | the server under test |
Neo4j or another graph database, loaded from passmcp-graph export --format cypher |
through the export | yes | write it yourself | through the export | the database's |
Evidence for each cell¶
- Reads passmcp's evidence directly.
ingestrecognises passmcp attestations andpassmcp check --output jsonreports (Ingesting evidence); a file it does not recognise is skipped and named. - No running service. The store is one JSON file,
.passmcp-graph/graph.json, read and written by the command that runs. - Policy gate.
check --policy FILEexits 1 and names every forbidden path, 0 when there is none, and 2 on an error (Policies). - Agents and identities. Agents come from client configurations and
identities from passmcp reports;
passmcp checkevaluates one server and does not record which agents use it. - Network.
TestNoCommandContactsTheNetworkrequires zero requests from every command (ADR 0001).
A graph database is the better choice for ad hoc exploration across data
passmcp does not produce; export --format cypher and --format graphml
exist so the two can be combined.