What is GNU Radio MCP (Marconi)?
GR-MCP, now renamed Marconi by its developer Yoel Bassin, is a Claude Code plugin with an MCP server and six skills for radio-frequency work in GNU Radio. The agent surveys a spectrum, finds and measures signals, builds a receiver pipeline, runs it, renders spectrograms and constellations, and saves the result as SigMF captures, YAML pipelines and .grc flowgraphs you can open in GNU Radio Companion.
The README calls it early development: v1.0 is simulation-only and does not touch hardware. It is GPL-3.0-or-later, written in Python, has about 57 GitHub stars, and its last commit, the v1.0 rewrite, landed on 13 June 2026.
How to install GNU Radio MCP (Marconi)
/plugin marketplace add yoelbassin/gr-mcp && /plugin install marconi (Claude Code)Is GNU Radio MCP (Marconi) worth it?
Marconi's design is better than its current reach. Its 19 tools cover a sensible arc: list_devices and capture, psd, find_signals, measure and detect_bursts for the survey; validate_pipeline, run_pipeline and export_grc for the receiver; spectrogram, psd_plot and constellation for the pictures. The six skills, such as build-receiver and debug-no-signal, tell the model how to chain them, and every run leaves files a person can inspect and rerun. Pipelines are validated before they run, which is the right instinct when a model is choosing filter parameters.
The limits are stated plainly in the project's own roadmap. Devices are simulated scenes, not radios; live hardware is planned for v1.2. The block vocabulary is analogue: it can demodulate FM but cannot recover bits from a digital signal, which the roadmap puts in v1.1 and which covers much of real RF work. Transmission exists only into a simulated scene and is gated by a confirmed=True flag that the skill itself calls a mistake-catcher, not a licensing check.
Two practical points. Installation is through the Claude Code plugin marketplace and needs uv, Python 3.13 or newer and a system-wide GNU Radio 3.10+, so other MCP clients have to start the server by hand from a clone; there is no PyPI package. And the escape-hatch skill tells the agent to write and run its own Python scripts with uv run when no tool fits, so the server can run code on your machine beyond its 19 tools. With 57 stars, one main contributor and no commit since June, it reads as a promising personal project rather than a tool to build a lab around.
Where it earns its place
A DSP student learning GNU Radio
See how an FM receiver is built, block by block
Ask Marconi to simulate an FM station at 100.3 MHz and build a receiver, then open the exported .grc flowgraph in GNU Radio Companion and compare it with the course notes. Free; needs GNU Radio 3.10+.
A radio amateur with a folder of IQ recordings
Find out what is in an unknown capture
Point the survey-spectrum skill at the file for a power spectrum, signal list and spectrogram. Analysing existing files does not need GNU Radio installed; demodulation stops at analogue signals.
Before you commit
- Setup
- In Claude Code: /plugin marketplace add yoelbassin/gr-mcp, then /plugin install marconi. Needs uv, Python 3.13+ and GNU Radio 3.10+ installed system-wide; uv starts the server on first use.
- Integrations
- GNU Radio backend; SigMF captures, YAML pipelines and .grc flowgraph export. Ships as a Claude Code plugin; other MCP clients can run the marconi-mcp stdio server from a clone.
- Data and privacy
- Runs locally with no network service. Captures and results go to your model provider. Artifacts are written under artifacts/ in the working directory or MARCONI_WORKSPACE. The escape-hatch skill writes and runs Python scripts on your machine.
- Cost at scale
- Free under GPL-3.0-or-later.
- Limits
- Simulation only in v1.0; analogue vocabulary with no digital decoding; transmit only into simulated scenes. Plugin manifest is still version 0.1.0.
What’s good
- 19 tools from spectrum survey to GNU Radio Companion export
- Pipelines validated before they run
- Leaves SigMF, YAML and .grc files you own
- Analysing existing IQ files needs no GNU Radio install
What’s not
- Simulation only; no real SDR hardware until v1.2
- No digital decoding until v1.1
- Escape-hatch skill has the agent run its own Python scripts
- Claude Code plugin install; no commit since 13 June 2026
Best for
Radio amateurs and DSP students who want an agent to prototype analogue receivers in simulation and hand back GNU Radio flowgraphs to study.
GNU Radio MCP (Marconi): common questions
- Is GNU Radio MCP (Marconi) worth it?
- A thoughtful start on letting an agent do RF work in GNU Radio, but until it can decode digital signals or drive a real radio it is a simulator with good manners. We rate it 3.2 out of 5.
- How much does GNU Radio MCP (Marconi) cost?
- Free and open source (GPL-3.0-or-later).
- What are the downsides of GNU Radio MCP (Marconi)?
- Simulation only; no real SDR hardware until v1.2; no digital decoding until v1.1; escape-hatch skill has the agent run its own Python scripts; claude Code plugin install; no commit since 13 June 2026.
- Who is GNU Radio MCP (Marconi) best for?
- Radio amateurs and DSP students who want an agent to prototype analogue receivers in simulation and hand back GNU Radio flowgraphs to study.
- How much does GNU Radio MCP (Marconi) cost for a team or in production?
- Free under GPL-3.0-or-later.
- Is GNU Radio MCP (Marconi) safe for company data?
- Runs locally with no network service. Captures and results go to your model provider. Artifacts are written under artifacts/ in the working directory or MARCONI_WORKSPACE. The escape-hatch skill writes and runs Python scripts on your machine.
Answers checked Nov 6, 2026.