Talk to your AI agent to run Simetrik. →
SimetrikDocs

Tools

Read tools, session tools and the Agent — how each behaves, and where the server stops.

Read tools, session tools, and the Agent

KindHow it behavesWhat it covers
ReadInstant, no confirmationWorkspace overview, resources, match rules, row data, dashboards, datasets, previews, lineage, the dependency graph — and collecting the Agent's reply.
SessionChanges reach, never dataAdd or drop a workspace from the current conversation.
AgentMinutes, confirms firstask_simetrik_agent — the only tool that builds or changes anything.

The read tools hit read-only endpoints: there's no server-side path from any of them to a write. That's the guarantee, rather than the annotation — MCP tool annotations are advisory, and clients are told not to treat them as a security boundary.

Building work — new sources, reconciliations, rule sets, dashboards — goes through ask_simetrik_agent, which hands the task to the Simetrik Agent. Because it changes things, your client asks before every run. You confirm once to start it; collecting the answer afterwards is a read, so waiting never prompts you again.

What the MCP server does, and what your agent does

The MCP server isn't a one-click "ask anything, get a finished reconciliation" tool. It's a bridge, and knowing where the line falls saves a lot of frustration.

The server does:

  • Return your real Simetrik configuration and data — resources, rules, rows, lineage.
  • Enforce your permissions on every call. It can't reach anything your own user can't.
  • Hand build-and-change work to the Simetrik Agent, which applies Simetrik's own methodology.

The server doesn't:

  • Decide what's correct for your operation. It reports what's configured; whether that's the right control is your call.
  • Upload local files. There's no filesystem here. Loading a file into a source needs the CLI or the web app.
  • Guarantee the phrasing of an answer. Your client's model writes the prose. The server supplies the facts.

On this page