Spaces:
Runtime error
Runtime error
Aster Vale commited on
Commit ·
f960cab
1
Parent(s): 5bcde7d
brand: tagline update
Browse files- signal-loom-mcp/README.md +188 -0
signal-loom-mcp/README.md
ADDED
|
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Signal Loom MCP Server
|
| 2 |
+
|
| 3 |
+
**Don't Transcribe it — Loom it.** Exposes [Signal Loom AI](https://signalloom.ai) transcription as native [Model Context Protocol (MCP)](https://modelcontextprotocol.io) tools. Works with Claude Desktop, AutoGen, and any MCP-compatible client.
|
| 4 |
+
|
| 5 |
+
*Loose threads become the lasting fabric in AI Context building.*
|
| 6 |
+
|
| 7 |
+
## Features
|
| 8 |
+
|
| 9 |
+
- **5 MCP tools**: `transcribe_audio`, `get_transcription_status`, `get_transcript_result`, `list_transcription_models`, `check_health`
|
| 10 |
+
- **Async stdio transport** — zero network ports required on the client side
|
| 11 |
+
- **Typed JSON Schema** tool definitions — works with any MCP client
|
| 12 |
+
- **Configurable via env vars** — no hardcoded URLs
|
| 13 |
+
|
| 14 |
+
## Requirements
|
| 15 |
+
|
| 16 |
+
- Python 3.10+ (the `mcp` package requires Python ≥3.10)
|
| 17 |
+
- Signal Loom API running at `http://localhost:18790` (or your configured URL)
|
| 18 |
+
|
| 19 |
+
## Install
|
| 20 |
+
|
| 21 |
+
```bash
|
| 22 |
+
pip install -e .
|
| 23 |
+
```
|
| 24 |
+
|
| 25 |
+
Or for a fast dev reinstall:
|
| 26 |
+
|
| 27 |
+
```bash
|
| 28 |
+
pip install -e ".[dev]"
|
| 29 |
+
```
|
| 30 |
+
|
| 31 |
+
## Configure
|
| 32 |
+
|
| 33 |
+
| Environment variable | Default | Description |
|
| 34 |
+
|---|---|---|
|
| 35 |
+
| `SIGNAL_LOOM_BASE_URL` | `http://localhost:18790` | Base URL of the Signal Loom API |
|
| 36 |
+
| `SIGNAL_LOOM_API_KEY` | _(none)_ | Optional API key for authenticated endpoints |
|
| 37 |
+
|
| 38 |
+
## Run
|
| 39 |
+
|
| 40 |
+
```bash
|
| 41 |
+
# Basic
|
| 42 |
+
python -m mcp_signalloom
|
| 43 |
+
|
| 44 |
+
# Or via the entry point (after pip install -e .)
|
| 45 |
+
mcp-signalloom
|
| 46 |
+
```
|
| 47 |
+
|
| 48 |
+
The server starts on **stdio** — connect your MCP client to the subprocess's stdin/stdout.
|
| 49 |
+
|
| 50 |
+
## Claude Desktop
|
| 51 |
+
|
| 52 |
+
Add to `~/Library/Application Support/Claude/claude_desktop_config.json`:
|
| 53 |
+
|
| 54 |
+
```json
|
| 55 |
+
{
|
| 56 |
+
"mcpServers": {
|
| 57 |
+
"signal-loom": {
|
| 58 |
+
"command": "python",
|
| 59 |
+
"args": ["-m", "mcp_signalloom"],
|
| 60 |
+
"env": {
|
| 61 |
+
"SIGNAL_LOOM_BASE_URL": "http://localhost:18790"
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
}
|
| 65 |
+
}
|
| 66 |
+
```
|
| 67 |
+
|
| 68 |
+
Restart Claude Desktop to pick up the new server. The `check_health` tool will appear in the tool list.
|
| 69 |
+
|
| 70 |
+
## AutoGen
|
| 71 |
+
|
| 72 |
+
```python
|
| 73 |
+
from autogen import ConversableAgent
|
| 74 |
+
|
| 75 |
+
agent = ConversableAgent(
|
| 76 |
+
"transcriber",
|
| 77 |
+
llm_config={
|
| 78 |
+
"tools": [
|
| 79 |
+
{
|
| 80 |
+
"type": "function",
|
| 81 |
+
"function": {
|
| 82 |
+
"name": "transcribe_audio",
|
| 83 |
+
"description": (
|
| 84 |
+
"Transcribe audio or video to structured JSON. "
|
| 85 |
+
"Supports MP3, WAV, M4A, FLAC, OGG, AAC, WebM, MP4, MOV."
|
| 86 |
+
),
|
| 87 |
+
"parameters": {
|
| 88 |
+
"type": "object",
|
| 89 |
+
"properties": {
|
| 90 |
+
"audio_url": {
|
| 91 |
+
"type": "string",
|
| 92 |
+
"description": "URL of the audio or video file to transcribe",
|
| 93 |
+
},
|
| 94 |
+
"language": {
|
| 95 |
+
"type": "string",
|
| 96 |
+
"description": "Language code (e.g. 'en', 'es', 'zh'). Auto-detected if omitted.",
|
| 97 |
+
},
|
| 98 |
+
"model": {
|
| 99 |
+
"type": "string",
|
| 100 |
+
"description": "Whisper model: mlx-community/whisper-large-v3-turbo (default), mlx-community/whisper-large-v3, openai/whisper-large-v3",
|
| 101 |
+
},
|
| 102 |
+
"return_format": {
|
| 103 |
+
"type": "string",
|
| 104 |
+
"enum": ["json", "text", "srt", "vtt"],
|
| 105 |
+
"description": "'json' (structured, default), 'text', 'srt', 'vtt'",
|
| 106 |
+
},
|
| 107 |
+
"word_timestamps": {
|
| 108 |
+
"type": "boolean",
|
| 109 |
+
"description": "Include word-level timestamps",
|
| 110 |
+
},
|
| 111 |
+
},
|
| 112 |
+
"required": ["audio_url"],
|
| 113 |
+
},
|
| 114 |
+
},
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"type": "function",
|
| 118 |
+
"function": {
|
| 119 |
+
"name": "get_transcription_status",
|
| 120 |
+
"description": "Check the status of a transcription job.",
|
| 121 |
+
"parameters": {
|
| 122 |
+
"type": "object",
|
| 123 |
+
"properties": {"job_id": {"type": "string"}},
|
| 124 |
+
"required": ["job_id"],
|
| 125 |
+
},
|
| 126 |
+
},
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"type": "function",
|
| 130 |
+
"function": {
|
| 131 |
+
"name": "get_transcript_result",
|
| 132 |
+
"description": "Get the completed transcript result for a finished transcription job.",
|
| 133 |
+
"parameters": {
|
| 134 |
+
"type": "object",
|
| 135 |
+
"properties": {"job_id": {"type": "string"}},
|
| 136 |
+
"required": ["job_id"],
|
| 137 |
+
},
|
| 138 |
+
},
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"type": "function",
|
| 142 |
+
"function": {
|
| 143 |
+
"name": "list_transcription_models",
|
| 144 |
+
"description": "List available Whisper models for transcription.",
|
| 145 |
+
"parameters": {"type": "object", "properties": {}},
|
| 146 |
+
},
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"type": "function",
|
| 150 |
+
"function": {
|
| 151 |
+
"name": "check_health",
|
| 152 |
+
"description": "Check if the Signal Loom API server is healthy and responding.",
|
| 153 |
+
"parameters": {"type": "object", "properties": {}},
|
| 154 |
+
},
|
| 155 |
+
},
|
| 156 |
+
],
|
| 157 |
+
},
|
| 158 |
+
)
|
| 159 |
+
|
| 160 |
+
# Agent can now call transcribe_audio(audio_url="https://example.com/podcast.mp3")
|
| 161 |
+
```
|
| 162 |
+
|
| 163 |
+
## MCP Inspector
|
| 164 |
+
|
| 165 |
+
```bash
|
| 166 |
+
# Start the Signal Loom API first (if running locally)
|
| 167 |
+
# Then start the MCP server in the background
|
| 168 |
+
python -m mcp_signalloom &
|
| 169 |
+
|
| 170 |
+
# Run the MCP inspector
|
| 171 |
+
npx @anthropic-ai/mcp-inspector
|
| 172 |
+
|
| 173 |
+
# Or connect via any MCP-compatible client
|
| 174 |
+
```
|
| 175 |
+
|
| 176 |
+
## Tools
|
| 177 |
+
|
| 178 |
+
| Tool | Description |
|
| 179 |
+
|---|---|
|
| 180 |
+
| `transcribe_audio` | Submit an audio/video URL for transcription. Returns a `job_id` (async) or the full transcript (sync, if ready). |
|
| 181 |
+
| `get_transcription_status` | Poll a job's status: `pending`, `processing`, `completed`, `failed`. |
|
| 182 |
+
| `get_transcript_result` | Retrieve the completed transcript JSON for a finished job. |
|
| 183 |
+
| `list_transcription_models` | List available Whisper models and their descriptions. |
|
| 184 |
+
| `check_health` | Ping the Signal Loom API health endpoint. |
|
| 185 |
+
|
| 186 |
+
## License
|
| 187 |
+
|
| 188 |
+
MIT
|