Build, deploy, test, and live-debug FAPs without ever touching the SD card. Three autonomous agents drive the device over USB RPC, decode crashes into source lines, and write the reproduction recipe for you.
Eight slash commands cover the loop end-to-end. Three agents take over when investigation needs autonomy. One MCP server holds the RPC session for both.
Recognizes OFW, Momentum, Unleashed, and RogueMaster. Routes deploys to the correct app directory; picks the matching ufbt channel automatically.
RPC storage handles the upload. No SD swap, no eject cycle. Survives device reboots if a crash interrupts the run.
128×64 framebuffer as PNG for visual fidelity and ASCII art for text assertions. Both are returned by every capture call.
Single press, long press, sequences with inter-step delays. Token grammar matches scenario YAML so muscle memory carries over.
RPC log subscription runs in the background. Crash signatures (BusFault, MemManage, furi_check) are tagged live as they arrive.
Auto-detects arm-none-eabi-addr2line via PATH or ufbt toolchain. Resolves PC/LR to source lines. Graceful fallback to heuristics.
Declarative, CI-friendly. Step kinds for launch, input, capture, assertion, storage, and on-failure hooks. Per-firmware gating.
Persistent RPC session shared by all agents. 25 native tools. Sub-100ms ping liveness check after every risky operation.
Hardware peripheral debugging on PA7/PA6/PA4/PB3/PB2/PC3/PC1/PC0. Configures pin mode and pull on the fly.
One Claude Code session, one persistent USB RPC session. Skills and agents share the MCP server's connection — no per-call reconnect cost.
Five commands. The first one detects everything else.
/connect-flipper // detect COM5, Momentum dev
/build-fap // ufbt, match channel
/deploy-fap // upload to /ext/apps/Tools/
/test-fap tests/basic.yaml // run, capture, assert
The MCP server gives agents persistent control. They reproduce the bug, capture state at each step, form hypotheses, and bisect until the root cause appears.
Reproduces a bug autonomously. Connects, deploys, navigates, narrows down, reports a reproduction recipe plus the suspected source line.
Adaptive YAML scenario runner. Adjusts mid-test when device behavior diverges. Classifies failures into CRASH / ASSERTION / UI / STORAGE / BUILD / LOGIC / FLAKE.
Decodes traces into source lines via addr2line. Ranks root-cause hypotheses with confidence levels. Optionally verifies the top guess by reproducing once on device.
Drop into ~/.claude/plugins/, restart Claude Code, run /connect-flipper. First-run bootstrap handles all Python deps automatically.
git clone https://github.com/halbothpa/flipper-dev-toolkit \
~/.claude/plugins/flipper-dev-toolkit
git clone https://github.com/halbothpa/flipper-dev-toolkit `
$HOME\.claude\plugins\flipper-dev-toolkit