SmartSDR MCP

A Model Context Protocol (MCP) server that lets AI assistants such as Claude control FlexRadio software-defined radios and operate CW (Morse code) QSOs semi-autonomously.

Overview

SmartSDR MCP bridges your FlexRadio hardware with any MCP-compatible AI assistant. The server exposes a set of tools and live resources that allow the AI to:

Discover and connect to FlexRadios on the local network
Control the radio — frequency, demodulation mode, and more
Decode Morse code in real-time using a custom DSP pipeline over DAX audio
Transcribe SSB voice using Whisper speech-to-text
Track QSO state — detect CQ calls, callsigns, signal reports, and exchanges
Generate contextually appropriate CW replies with AI assistance
Transmit safely with a proposal-approval loop: the AI suggests, you approve, then it transmits
Contest automation (experimental) — monitor SSB contest frequencies, identify stations, and auto-call
Band Scout — real-time band condition monitoring from DX spot analysis
Auto-DX Hunter — cross-reference spots against your logbook to find unworked DXCC entities

The design keeps humans in the loop: no RF is ever emitted without explicit approval.

Features

Radio Control

Discover all FlexRadios on the local network (or target a specific serial number or IP)
Connect and disconnect from the radio programmatically
Read current frequency, mode, TX state, and CW pitch
Set VFO frequency (e.g., 14.035 MHz for 20 m CW)
Set demodulation mode: CW, USB, LSB, AM, FM
List and configure all receiver slices (frequency, mode, filter, antenna, audio, etc.)
Adjust receiver passband filters, RIT/XIT offsets, and AGC settings
Control noise reduction (NR, NB, ANF, WNB) with individual levels
Read real-time meters: S-meter, SWR, forward/reflected power, PA temperature, voltage, ALC
Set RF transmit power and tune power levels
Manage Tracking Notch Filters (TNF) to null out interference
Control antenna tuner (ATU): status and auto-tune
Save, load, and delete memory channels
View and manage DX spots
Read GPS/GNSS data: grid square, coordinates, altitude, satellites
Select RX and TX antennas per slice
Control per-slice audio: gain, pan, mute
Direct TX control: PTT (MOX), tune carrier, TX monitor, TX inhibit
8-band TX/RX audio equalizer (63Hz–8kHz)
Microphone settings: level, boost, bias, input source
TX audio processing: compander and speech processor
VOX (Voice-Operated Transmit): enable, level, delay
Panadapter control: center frequency, bandwidth, dBm range, FPS
Tune to DX spot by callsign

CW (Morse Code) Decoding

Captures audio via the FlexRadio DAX (Digital Audio eXchange) interface
Custom DSP pipeline:
- Band-pass filter tuned to the radio's configured CW pitch
- AGC (automatic gain control)
- Envelope detector and key-event extractor
- WPM estimator (auto-detect or fixed)
- Ham-aware rescoring to validate callsigns and common CW prosigns
Real-time decode buffer with diagnostic telemetry (WPM, tone magnitude, noise floor)
Decoded message history with timestamps

SSB Voice Transcription

Real-time speech-to-text using Whisper (small.en or base.en models)
Ham radio contest-optimized prompt for better callsign/phonetic recognition
Hallucination filtering to suppress common Whisper artifacts
Live transcription buffer and segment history

QSO Tracking

State machine that understands the CW QSO lifecycle:
- Idle → Calling CQ / Answering CQ → Exchange → Closing → Idle
Detects CQ calls, called/calling callsigns, signal reports (RST), and exchange fields
Tracks your callsign against decoded traffic

AI-Assisted Replies

CwGenerateReply produces a contextual CW response based on current QSO state
Returns a proposal (ID, suggested text, reason, estimated WPM, estimated duration)
Proposals sit in a pending queue until explicitly approved

Safe Transmission

CwSendText approves and sends a pending proposal or transmits custom freeform text
Emergency abort (CwAbort) stops transmission immediately
QSO tracker is notified of every sent message to keep state consistent

Contest Agent (Experimental)

Monitors SSB contest frequencies via Whisper transcription + Claude Haiku analysis
Identifies running stations calling CQ and determines optimal call timing
DX cluster integration (HamQTH) for station identification when transcription is ambiguous
Auto-call mode with TTS voice TX through DAX TX
State machine tracks QSO progression: Monitoring → CallingStation → ExchangingReports → Completing

Note: Voice TX audio quality is experimental. DAX TX buffer pacing can cause distortion under some conditions.

Band Scout

Real-time HF band condition monitoring using DX spots from the radio
Scans spots every 15 seconds and groups activity by band (160m–6m)
Reports per-band: spot count, active modes, unique callsigns, frequency range, and activity score
Helps identify which bands are open and active right now

Auto-DX Hunter

Loads your ADIF logbook to determine which DXCC entities you've already worked
Monitors live DX spots and cross-references against your log
Identifies unworked DXCC entities by band and mode
Includes a ~200-entity DXCC prefix lookup table covering ITU allocations and special prefixes
One-click QSY to the highest-priority unworked entity
Mark entities as worked on the fly without reloading the log

Live MCP Resources

Resource URI	Description
`flex://radio/state`	Current radio state as JSON
`flex://cw/live`	Real-time CW decode buffer with diagnostics
`flex://cw/recent`	Last 20 decoded messages
`flex://cw/proposals`	Pending AI reply proposals awaiting approval

Prerequisites

Requirement	Details
.NET 9 SDK	Download
FlexRadio hardware	Flex 6000/8000 series on the same network subnet
SmartSDR (optional)	Running for GUI access; required for DAX audio to be available
MCP client	Claude Desktop, or any MCP-compatible host
Whisper model (for SSB)	`ggml-small.en.bin` or `ggml-base.en.bin` in `models/` directory
Anthropic API key (for contest agent)	Set `ANTHROPIC_API_KEY` env var for Claude Haiku situation analysis

Installation

# 1. Clone the repository
git clone https://github.com/patrickrb/smartsdr-mcp.git
cd smartsdr-mcp

# 2. Restore NuGet packages
dotnet restore

# 3. Build the solution
dotnet build --configuration Release

Configuration

Open src/SmartSdrMcp/Program.cs and update the constants near the top:

const string MyCallsign = "K1AF";    // ← your amateur radio callsign
const string MyName     = "PATRICK"; // ← your name (used in CW exchanges)
const string MyQth      = "Kansas";  // ← your location (used in contest exchanges)

These values are used by the QSO tracker (to recognize messages addressed to you) and by the AI reply generator (to sign outgoing CW).

No other configuration files are required. All runtime parameters (DAX channel, WPM lock, frequency) are passed as tool arguments.

Running the Server

The MCP server communicates over stdio, so it is launched by your MCP client automatically.

For manual testing:

dotnet run --project src/SmartSdrMcp/SmartSdrMcp.csproj

Claude Desktop integration

Add the following entry to your claude_desktop_config.json:

{
  "mcpServers": {
    "smartsdr": {
      "command": "dotnet",
      "args": [
        "run",
        "--project",
        "/absolute/path/to/smartsdr-mcp/src/SmartSdrMcp/SmartSdrMcp.csproj"
      ]
    }
  }
}

Or, if you prefer to run the compiled binary:

{
  "mcpServers": {
    "smartsdr": {
      "command": "/absolute/path/to/smartsdr-mcp/src/SmartSdrMcp/bin/Release/net9.0/SmartSdrMcp"
    }
  }
}

MCP Tools Reference

Radio Tools

Tool	Parameters	Description
`connect_radio`	`serial?`, `ip?`	Discover and connect to a FlexRadio on the network
`list_radios`	—	Discover radios currently visible on the network
`disconnect_radio`	—	Disconnect from the currently connected radio
`get_radio_state`	—	Get current frequency, mode, TX state, and CW pitch
`radio_health`	—	Get overall radio/listener health diagnostics
`set_frequency`	`frequencyMHz`	Set the active slice frequency in MHz (e.g., `14.035`)
`step_frequency`	`hz`	Step active slice frequency by N Hz (positive or negative)
`set_active_slice`	`index`	Set the active slice by index
`set_mode`	`mode`	Set the demodulation mode: `CW`, `USB`, `LSB`, `AM`, or `FM`
`set_cw_profile`	`wpm?`, `pitch?`, `breakIn?`, `iambic?`	Set CW profile values in one operation
`get_meters`	—	Get real-time meter readings (S-meter, SWR, power, ALC, voltage, PA temp, mic, compression)
`get_filter`	—	Get current receiver passband filter bounds (low/high Hz)
`set_filter`	`low`, `high`	Set receiver passband filter bounds in Hz
`set_rit`	`enabled?`, `offsetHz?`	Set RIT (Receiver Incremental Tuning)
`set_xit`	`enabled?`, `offsetHz?`	Set XIT (Transmitter Incremental Tuning)
`list_slices`	—	List all receiver slices with full configuration
`get_noise_reduction`	—	Get noise reduction settings (NR, NB, ANF, WNB)
`set_noise_reduction`	`nrOn?`, `nrLevel?`, `nbOn?`, `nbLevel?`, `anfOn?`, `anfLevel?`, `wnbOn?`, `wnbLevel?`	Set noise reduction on the active slice
`get_gps`	—	Get GPS data: lat, lon, grid square, altitude, satellites
`list_tnfs`	—	List all Tracking Notch Filters
`add_tnf`	`frequencyMHz`	Add a Tracking Notch Filter at a frequency
`remove_tnf`	`tnfId`	Remove a Tracking Notch Filter by ID
`get_rf_power`	—	Get current RF power, tune power, and max power level
`set_rf_power`	`rfPower?`, `tunePower?`	Set RF transmit power (0-100) and/or tune power
`get_atu_status`	—	Get antenna tuner status (present, enabled, tuning, bypass)
`atu_tune`	—	Initiate ATU auto-tune cycle
`list_memories`	—	List all saved memory channels
`load_memory`	`index`	Load (recall) a saved memory channel
`delete_memory`	`index`	Delete a saved memory channel
`list_spots`	—	List all DX spots with callsign, frequency, mode, spotter
`remove_spot`	`callsign`	Remove a DX spot by callsign
`get_agc`	—	Get AGC settings (mode, threshold, off-level)
`set_agc`	`mode?`, `threshold?`, `offLevel?`	Set AGC mode (off/slow/medium/fast) and threshold
`get_antenna`	—	Get current RX/TX antenna and available antenna lists
`set_antenna`	`rxAnt?`, `txAnt?`	Set RX and/or TX antenna for the active slice
`get_slice_audio`	—	Get audio gain, pan, and mute for the active slice
`set_slice_audio`	`audioGain?`, `audioPan?`, `mute?`	Set slice audio gain (0-100), pan (0-100), mute
`get_tx_state`	—	Get TX state: MOX, TX tune, TX monitor, TX inhibit
`set_tx`	`txMonitor?`, `txInhibit?`	Set TX monitor and safety inhibit (MOX/tune via guarded TX controls)
`get_equalizer`	`select`	Get TX or RX equalizer settings and band levels
`set_equalizer`	`select`, `enabled?`, `hz63?`–`hz8000?`	Set equalizer bands (dB) for TX or RX
`get_mic`	—	Get mic level, boost, bias, input source
`set_mic`	`micLevel?`, `micBoost?`, `micBias?`, `micInput?`	Set microphone settings
`get_tx_audio_processing`	—	Get compander and speech processor settings
`set_tx_audio_processing`	`companderOn?`, `companderLevel?`, `speechProcessorEnable?`, `speechProcessorLevel?`	Set TX audio processing
`get_vox`	—	Get VOX settings: enabled, level, delay
`set_vox`	`enabled?`, `level?`, `delay?`	Set VOX enabled, level, and delay
`list_panadapters`	—	List all panadapters with center freq, bandwidth, dBm range
`set_panadapter`	`streamId`, `centerFreq?`, `bandwidth?`, `lowDbm?`, `highDbm?`, `fps?`, `average?`	Set panadapter display settings
`tune_to_spot`	`callsign`	Tune active slice to a DX spot's frequency

CW Listener Tools

Tool	Parameters	Description
`cw_listener_start`	`daxChannel` (default `1`), `fixedWpm` (default `0` = auto)	Start Morse decoding on the specified DAX channel
`cw_listener_stop`	—	Stop CW decoding and release the DAX audio stream
`cw_get_live_text`	—	Get the real-time decode buffer with diagnostics
`cw_decode_diagnostics`	—	Get low-level decoder diagnostics for troubleshooting
`cw_get_recent_messages`	`count` (default `10`)	Get the last N decoded messages with timestamps
`cw_get_qso_state`	—	Get the current QSO state machine snapshot
`cw_reset_qso`	—	Reset QSO tracking back to Idle
`record_audio_start`	`format`, `seconds?`	Start recording DAX audio (wav or raw)
`record_audio_stop`	—	Stop recording and return file details
`qso_export`	`format`	Export QSO data (json or adif)

CW Transmit Tools

Tool	Parameters	Description
`cw_generate_reply`	—	Generate an AI-suggested CW reply based on QSO state
`cw_get_pending_replies`	—	List all proposals currently awaiting approval
`clear_pending_replies`	—	Clear all queued reply proposals
`set_tx_guard`	`armed?`, `maxSeconds?`, `requireProposal?`	Configure TX guard safety settings
`get_tx_guard`	—	Get current TX guard settings
`cw_send_text`	`proposalId?`, `customText?`, `wpm` (default `20`)	Approve and transmit a pending proposal, or send custom text
`cw_abort`	—	Emergency stop — immediately halt CW transmission
`run_macro`	`name`	Run predefined macro workflows (find_cq, answer_cq, close_qso_safely)

SSB Listener Tools

Tool	Parameters	Description
`ssb_listener_start`	`daxChannel` (default `1`)	Start SSB speech-to-text listener using Whisper
`ssb_listener_stop`	—	Stop SSB speech-to-text listener
`ssb_get_live_text`	—	Get live SSB transcription
`ssb_get_recent_messages`	`count` (default `10`)	Get the last N transcribed SSB speech segments

Contest Agent Tools (Experimental)

Tool	Parameters	Description
`contest_agent_start`	`callsign`, `name?`, `qth?`, `autoMode?`	Start the contest agent (set `ANTHROPIC_API_KEY` env var for Haiku analysis)
`contest_agent_stop`	—	Stop the contest agent
`contest_agent_status`	—	Get current agent status (stage, prompt, running station, QSO count)
`contest_agent_ack`	—	Acknowledge current prompt (operator has spoken)
`contest_agent_skip`	—	Skip current opportunity, return to monitoring
`contest_agent_log`	—	Get list of completed contest QSOs
`contest_voice_test`	`text` (default `"Kilo One Alpha Foxtrot"`)	Test voice TX with TTS or a 1kHz tone (`text="tone"`) — requires TX guard armed

Band Scout Tools

Tool	Parameters	Description
`band_scout_start`	—	Start band condition monitoring (scans DX spots every 15s)
`band_scout_stop`	—	Stop band condition monitoring
`band_scout_report`	—	Get current band activity report with per-band spot counts, modes, and activity scores

Auto-DX Hunter Tools

Tool	Parameters	Description
`dx_hunter_load_log`	`adifFilePath`	Load an ADIF log file to identify already-worked DXCC entities
`dx_hunter_start`	`band?`, `mode?`, `autoTune?`	Start monitoring DX spots for unworked DXCC entities (optionally limited by band/mode and with optional auto-tuning)
`dx_hunter_stop`	—	Stop DX spot monitoring
`dx_hunter_status`	—	Get hunter status: running state, entities worked, needs count, last scan time
`dx_hunter_needs`	—	List unworked DXCC entities currently spotted
`dx_hunter_tune_next`	—	QSY to the highest-priority unworked DXCC spot
`dx_hunter_mark_worked`	`callsign`, `band?`, `mode?`	Mark a DXCC entity as worked (by callsign, with optional band and mode)
`dx_hunter_lookup`	`callsign`	Look up the DXCC entity for a callsign

Typical Workflow

1.  connect_radio()              – discover and connect to the radio
2.  set_frequency(14.035)        – tune to 20 m CW
3.  set_mode("CW")               – set demodulation mode
4.  cw_listener_start()          – begin decoding Morse
         ┌─ cw_get_live_text()   – monitor real-time decode
         └─ cw_get_recent_messages() – review decoded history
5.  cw_get_qso_state()           – check what the AI knows about the QSO
6.  cw_generate_reply()          – AI proposes a reply (no RF yet)
7.  cw_get_pending_replies()     – review the proposal
8.  cw_send_text(proposalId=...) – approve → transmit
         or
    cw_abort()                   – cancel at any time
9.  cw_listener_stop()           – done for the session
10. disconnect_radio()

Architecture

┌───────────────────────────────────────────────────┐
│               MCP Client (e.g. Claude)            │
└──────────────────────┬────────────────────────────┘
                       │ stdio (MCP protocol)
┌──────────────────────▼────────────────────────────┐
│                  SmartSDR MCP Server               │
│                                                   │
│  Tools ──────────────────────── Resources         │
│  RadioTools          flex://radio/state           │
│  CwListenerTools     flex://cw/live               │
│  CwTransmitTools     flex://cw/recent             │
│  SsbListenerTools    flex://cw/proposals          │
│  ContestTools                                     │
│  BandScoutTools                                   │
│  DxHunterTools                                    │
│                                                   │
│  Core Services                                    │
│  RadioManager   ←→  FlexLib (FlexRadio API)       │
│  AudioPipeline  ←→  DAX (24 kHz PCM over network) │
│  CwPipeline         BPF → Goertzel → Envelope     │
│  SsbPipeline        Whisper speech-to-text        │
│  ContestAgent       Haiku analysis + state machine│
│  QsoTracker         State machine (QSO lifecycle) │
│  ReplyGenerator     Contextual CW reply logic     │
│  TransmitController Proposal queue + TX safety    │
│  VoiceTransmitter   TTS → DAX TX streaming        │
│  ClusterClient      DX cluster spot lookups       │
│  BandScoutMonitor   DX spot band activity scoring │
│  DxHunterAgent      DXCC needs vs. live spots     │
│  DxccLookup         Callsign → DXCC entity        │
└──────────────────────┬────────────────────────────┘
                       │ FlexLib / VITA 49
┌──────────────────────▼────────────────────────────┐
│          FlexRadio Hardware (Flex 6000/8000)       │
└───────────────────────────────────────────────────┘

Key Design Principles

Proposal-approval loop — the AI suggests CW text; a human must explicitly approve before any RF is emitted.
Event-driven pipeline — audio samples flow through DSP stages and fire events consumed by higher-level services.
Single-responsibility services — each layer (audio, DSP, segmentation, QSO state, AI, TX) is independently testable and replaceable.
Stdio transport — the server has no open ports; it is driven entirely by the MCP client over standard I/O.

Tech Stack

Component	Details
Language	C# 12 with nullable reference types
Framework	.NET 9.0
MCP library	`ModelContextProtocol` v1.0.0
Radio API	FlexLib (ported to .NET 9, included in `lib/`)
Speech-to-text	Whisper.net (whisper.cpp bindings)
AI analysis	Anthropic Claude Haiku (contest agent)
Data streaming	VITA 49 protocol (UDP) for real-time audio and waterfall
Audio	24 kHz mono PCM de-interleaved from DAX stereo stream
DI / hosting	`Microsoft.Extensions.Hosting` v9.0.0

License

See LICENSE for details.

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
lib/FlexLib_Core		lib/FlexLib_Core
src		src
.gitignore		.gitignore
README.md		README.md
SmartSdrMcp.sln		SmartSdrMcp.sln

Folders and files

Latest commit

History

Repository files navigation

SmartSDR MCP

Overview

Features

Radio Control

CW (Morse Code) Decoding

SSB Voice Transcription

QSO Tracking

AI-Assisted Replies

Safe Transmission

Contest Agent (Experimental)

Band Scout

Auto-DX Hunter

Live MCP Resources

Prerequisites

Installation

Configuration

Running the Server

Claude Desktop integration

MCP Tools Reference

Radio Tools

CW Listener Tools

CW Transmit Tools

SSB Listener Tools

Contest Agent Tools (Experimental)

Band Scout Tools

Auto-DX Hunter Tools

Typical Workflow

Architecture

Key Design Principles

Tech Stack

License

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages