RF LAB.

From a tool call to a trustworthy operation.

A reliability-focused bridge between CST Studio Suite and MCP clients, covering connection timing, error propagation, and the simulation lifecycle.

The source repository may open on an external site and can require private access.

MCP clientValidated tool callCST session and results
Architecture flow. A successful tool call is not the same as a completed and validated simulation.
Reliability fixes merged; 56 tests passed on each of two MCP SDK versions.12 September 2026 · Software project

A server that reports operation state accurately.

The review found CST connections occurring before the client handshake, some failures being returned as successful calls, and inconsistent solver lifecycle behavior.

Client connection

Standard MCP handshake, tool discovery, and calls, with the stdio protocol channel kept separate from logs.

CST lifecycle

Blocking and asynchronous solver calls are distinguished, while status queries remain read-only.

Actionable errors

Failures reach the client as failures, and an ambiguous operation state is preserved when a timeout occurs.

What was verified.

  • 56 tests passed with MCP SDK 1.29.0 and 56 with 2.2.0; 180 tool schemas were validated.
  • Handshake, tool discovery, calls, and separation of Python/native logs were tested through a real stdio subprocess.
  • A fake CST backend verified busy, timeout, unknown solver state, safe disconnect, and result-error propagation.

Open limits.

  • All 180 tools have not been executed against a real CST installation, and correctness is not claimed for every client or CST release.
  • The running research experiment was preserved; this work did not run a new end-to-end solve in real CST.

Before first use.

For a new setup, add the MCP server first and then connect explicitly with cst_connect. A real-CST check should later use an empty session and a temporary project. The verified scope is recorded in docs/RELIABILITY_REVIEW.md in the repository.