🌐 English · Русский
MCP server that gives Claude Code access to your local Froggy context — screen, meetings, local LLM.
Froggy runs on your Mac: it sees your screen via OCR,
transcribes calls, and runs a local LLM on 8 GB RAM. Claude Code is a powerful cloud model but
has no eyes. froggy-mcp connects them — Claude can query Froggy over its Unix socket without
any copy-paste.
Status: personal-use companion to Froggy. Not a product. Requires the Froggy daemon running locally.
📬 Contact: @froggychips on Telegram 📜 License: MIT
| Repo | Role |
|---|---|
| Froggy | Local LLM daemon, OCR, screen context, memory management |
| FroggyKit | Shared Swift package — FroggyClient IPC |
| froggy-mcp | MCP bridge → Claude Code |
| froggy-sre | SRE incident response agent — K8s alert analysis pipeline |
Claude Code ←—stdio / JSON-RPC—→ froggy-mcp ←—unix socket / JSON-line—→ Froggy daemon
Claude Code ←—stdio / JSON-RPC—→ froggy-sre ←—socket (primary) / HTTPS (fallback)—→ Froggy / Anthropic
| Tool | What it does |
|---|---|
froggy_status |
Daemon status: model loaded, memory pressure, audio devices, transcription state |
froggy_context |
Latest OCR snapshot of your screen — what's open right now |
froggy_generate |
Generate via local LLM (private, no cloud) |
froggy_transcript |
Transcript of the current or last call (markdown, with timestamps) |
froggy_speak |
Speak text aloud via macOS system TTS |
froggy_freeze |
Freeze an app (SIGSTOP) to free unified memory before heavy tasks |
froggy_thaw_all |
Unfreeze all apps (SIGCONT) after a heavy task completes |
froggy_pressure |
Current memory pressure level and list of frozen processes |
froggy_listen |
Start call recording (microphone transcription); optionally inject pre-call context |
froggy_listen_stop |
Stop call recording; returns session path for recap |
froggy_recap |
Generate an LLM summary of the last call via the local model |
froggy_inject |
Inject text (Jira tickets, notes) into the current call's session file |
froggy_chat |
Fast voice reply via local LLM — no roundtrip to Claude API; ~1–2 s latency |
- macOS 14+
- Froggy daemon running
# 1. Build the binary
git clone https://github.com/froggychips/froggy-mcp
cd froggy-mcp
swift build -c release
cp .build/release/froggy-mcp /usr/local/bin/froggy-mcp
# 2. Register with Claude Code
claude mcp add froggy /usr/local/bin/froggy-mcp --scope globalAfter restarting Claude Code the froggy_* tools will be available in every session.
Connects to ~/Library/Application Support/Froggy/froggy.sock by default.
Override: FROGGY_IPC_SOCKET=/path/to/froggy.sock.
froggy-mcp is a minimal Swift binary. It speaks
MCP (JSON-RPC 2.0 over stdio) toward Claude Code and
speaks Froggy's Unix-socket IPC toward the daemon via FroggyKit.
The bridge is ~350 lines of Swift.
Claude Code ←—stdio / JSON-RPC—→ froggy-mcp ←—unix socket / JSON-line—→ Froggy daemon
Part of the Froggy ecosystem.