Skip to content

MCP Overview

RFlect ships a Model Context Protocol server so an AI agent (Claude Code, Cline, Continue, etc.) can drive RFlect programmatically. No GUI required.

RFlect has no LLM of its own

As of v5.0.0 RFlect makes no outbound LLM/API calls and needs no API key or subscription. It is a deterministic RF analysis + rendering toolkit. The driving MCP agent is the LLM: it calls RFlect's tools for data. If you want narrative prose in a report, the agent authors it and passes it to generate_report via the narrative parameter.

What you can do over MCP

  • Import passive HPOL/VPOL pairs, active TRP files, S-parameter sweeps
  • Run analysis: HPBW, F/B ratio, gain stats, polarization, UWB SFF, impedance bandwidth, S11/VSWR, group delay
  • Compare antennas head-to-head; estimate link budget / range; compute MIMO diversity
  • Generate DOCX reports (deterministic prose by default, or agent-authored)
  • Generate active chamber calibration files and track calibration drift across time
  • Run a standard procedure on a folder with a single call via process_folder: auto-detects intent, runs the right workflow, optionally generates a report
  • Optional instrument control (VNA / positioner) against an in-memory mock, or live hardware if the instruments extra is installed

Tool count

61 tools. See Tools Reference for the list with signatures, and MCP_STATUS.md for the category table.

Smoke check after install: tools: 61 from the snippet in Installation.

Why an orchestrator?

Before v4.2.0, an MCP client wanting to "process this folder and give me a report" had to:

  1. list_measurement_files(folder) to see what's inside
  2. Decide passive vs active based on filenames
  3. bulk_process_passive(folder, freqs) or bulk_process_active(folder)
  4. import_passive_pair(...) or import_active_processed(...) for each one
  5. generate_report(path, options)

That's a five-step chain that every agent had to script. process_folder collapses all of it into one call:

process_folder("/path/to/captures", intent="auto", report=True)

See Recipes for the full set of standard procedures.

Where to next