anythingllm-mcp
# AnythingLLM MCP Server
Complete MCP (Model Context Protocol) server for AnythingLLM - the all-in-one AI document chat platform.
## Features
Zero human interaction. Just set your API key and go.
### Tools (15 total)
**System**
- `check_token` - Validate API token
- `generate_api_key` - Generate new API key (admin)
**Workspaces**
- `list_workspaces` - List all workspaces
- `get_workspace` - Get workspace details
- `create_workspace` - Create new workspace
- `update_workspace` - Update workspace settings
- `delete_workspace` - Delete a workspace
**Chat**
- `chat` - Send chat message (chat/query mode)
- `stream_chat` - Stream chat response
**Documents**
- `upload_document` - Upload document to workspace
- `update_embeddings` - Manage workspace embeddings
- `list_workspace_documents` - List workspace documents
**Threads**
- `list_threads` - List workspace threads
- `get_thread` - Get thread details
- `delete_thread` - Delete a thread
**System & Compatible**
- `get_system_env` - Get system configuration
- `openai_chat_completion` - OpenAI-compatible endpoint
## Quick Start
### 1. Install
```bash
npm install -g anythingllm-mcp
```
### 2. Set API Key
Get your API key from AnythingLLM: **Settings > Developer API**
```bash
export ANYTHING_LLM_API_KEY="your-api-key-here"
export ANYTHING_LLM_BASE="http://localhost:3001/api/v1" # optional, default
```
### 3. Add to OpenCode
Add to your `opencode.json`:
```json
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"anythingllm": {
"type": "local",
"command": ["anythingllm-mcp"],
"environment": {
"ANYTHING_LLM_API_KEY": "your-api-key-here",
"ANYTHING_LLM_BASE": "http://localhost:3001/api/v1"
},
"enabled": true
}
}
}
```
### 4. Use in Any Session
The tools are now available. Example:
```
> Use the chat tool to ask AnythingLLM about my documents
```
## Development
```bash
git clone https://github.com/moliv/anythingllm-mcp.git
cd anythingllm-mcp
npm install
npm run build
```
## Requirements
- AnythingLLM running (local or remote)
- Valid API key from AnythingLLM
- Node.js 18+
## License
MIT
TDQS
Scored across 17 tools
All tools target distinct operations: chat, workspace, thread, document, and admin functions. Chat variants (chat, stream_chat, openai_chat_completion) differ by mode, and workspace/thread/document tools have clear boundaries.
Most tools follow verb_noun pattern in snake_case (e.g., create_workspace, list_threads). A few like 'chat' and 'openai_chat_completion' deviate slightly, but overall naming is predictable.
17 tools cover workspace, document, thread, chat, and admin operations without feeling excessive. The scope matches the server's purpose well.
CRUD operations are present for workspaces and threads (though thread creation is implicit via chat). Document management includes upload, listing, and embedding updates. Missing explicit delete document but covered via update_embeddings.