reference faq

Reference and FAQ

Which interface should a new client use?

Use the exact-schema ida_* operations. The old broad tool(action=...) interface remains for compatibility and is selected with IDA_MCP_TOOL_SURFACE=legacy.

Where do I get an operation's exact arguments?

Ask the server:

ida_help(query="ida_decompile")

You can also consult the generated operation reference. This wiki intentionally focuses on workflows rather than repeating every schema.

What does risk_ack mean?

It is an explicit acknowledgement required by IDB-writing operations and other destructive session actions. It does not override operator policy. Set it only after checking the target, intended change, and rollback plan.

Does analysis data leave the machine?

The server itself has no LLM service behind it and uses deterministic local IDA calls. Local embedding and reranking stay on the machine. The Gemini embedding backend is opt-in and uploads compact behavioral signatures to Google, not full decompilations.

Are embeddings required?

No. ida_find, listings, code inspection, findings, and ordinary session operations work without semantic models. Install models only when behavioral search or reranking is useful.

What is safe mode?

Safe mode protects whole-binary operations while IDA auto-analysis is pending. Small-area manual reads and several annotation operations remain available. Wait for ida_session_status to confirm analysis completion before indexing or running other gated work.

How do findings differ from IDB comments?

Findings live in the persistent investigation workspace and retain lifecycle, confidence, evidence, conflicts, and stale state. Publishing copies reviewed confirmed conclusions into the IDB. Importing annotations adopts existing names and comments back into the workspace.

How do I recover from a bad edit?

Use ida_idb_snapshot before experiments and ida_idb_restore_snapshot to roll back. Undo transactions can bracket a batch. Patches and undefines should be treated as destructive; verify bytes and snapshots before using them.

What does Alpha mean here?

The project is Alpha (1.0.0a1), so operation schemas and behavior may move. Pin a commit when reproducibility matters.

Is the optional threat corpus installed automatically?

No. It is opt-in with the installer. Normal installs do not download it.

Where are policy settings documented?

The operator baseline uses IDA_MCP_POLICY_MODE or ~/.config/ida-pro-mcp/policy.json. See the safety model.

References: README.md, generated operation reference.