The Pathix MCP server serves the parsed dependency and security graph to any HTTP-native client: Claude Desktop, Claude Code, Cursor. The agent signs in as you, sees what you can see, and cannot change a thing.
The question Pathix gets asked most is some version of this: we shipped a solution on Thursday and a field started behaving differently on Friday. In the console that is three screens. Over MCP the agent does the narrowing itself.
What comes back is the same data the console shows, tier included. The agent can tell you which of those six writers is pinned by a registration and which one a model read out of compiled code, because that distinction travels with the edge rather than being flattened on the way out.
Each one returns structured data rather than prose, so the agent composes them instead of parsing paragraphs. They are grouped here by the job they belong to.
Where the agent starts when it knows nothing about the environment yet.
Turn a half-remembered name into the actual component, then get its detail.
The core question, in both directions, plus everything bound to a component or a table.
Who can actually reach a column, resolved through roles, teams, business units and field security.
The verdict lane: deterministic checks that fired, with their severity.
What one operation reaches, and what order to rewrite things in.
Handing an agent a tool that reaches production is the part a security lead will want to talk about. Three properties settle most of that conversation.
No tool in the surface modifies your Dynamics environment. None of them changes triage state inside Pathix either, so an agent cannot quietly close a finding or dismiss an AI edge on your behalf. Reading is the whole contract.
The agent uses your Entra identity, not a service account. It sees the environments and components you can already see, and connecting an agent widens nobody's access, which is usually the first question a security lead asks.
Agents get structure: schema, registrations, definitions, roles, and the edges between them. Record values were never stored in Pathix, so there is nothing of that kind for an agent to retrieve.
The OAuth client registers itself the first time a client connects, so there is no app registration to provision before anyone can try it. A wizard in Settings emits the config block for the client you are using, and setup is copying that block across.
Because the server speaks plain MCP over HTTP, a client nobody has heard of yet works the same way as the ones named above.
Microsoft's Dataverse MCP hands an agent your data, your schema, and record operations, and it is well built for that. It does not parse component logic into column-level dependency edges, so an agent holding only that surface can see a plugin step is registered and still not know which column it writes. Pathix answers the second question. Most agents that need both end up holding both.
The demo is the real console running on a sample environment. A walkthrough can take the same questions through an agent instead.