> ## Documentation Index > Fetch the complete documentation index at: https://docs.60db.ai/llms.txt > Use this file to discover all available pages before exploring further. # Introduction # MCP Server The 60db MCP (Model Context Protocol) Server enables AI assistants like Claude to directly interact with the full 60db platform — voice AI, persistent memory, and access-control checks. **Source code:** [github.com/60db-ai/mcp](https://github.com/60db-ai/mcp) · **npm:** [`60db-mcp-server`](https://www.npmjs.com/package/60db-mcp-server) · **License:** MIT ## What is the MCP Server? The MCP Server provides a standardized interface for AI assistants to: * **Generate speech** from text using any voice * **Transcribe audio** files with speaker diarization * **Create custom voices** from audio samples * **Manage workspaces** and team collaboration * **Store and recall memories** — persistent memory and RAG across sessions * **Upload documents** — PDF, DOCX, XLSX, PPTX, EML, scanned images with built-in OCR * **Assemble LLM-ready context** — purpose-built for retrieval-augmented generation * **Check permissions** — query Cerbos policy before attempting gated actions * **Track usage** and analytics * **View monthly spend** across every billable operation ## Quick Start ### 1. Install the Server ```bash theme={null} # From npm npm install -g 60db-mcp-server # Or clone from source git clone https://github.com/60db-ai/mcp.git cd mcp npm install npm run build ``` ### 2. Configure Claude Desktop Add to your Claude Desktop config: **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json` **Windows**: `%APPDATA%/Claude/claude_desktop_config.json` ```json theme={null} { "mcpServers": { "60db": { "command": "node", "args": ["/path/to/mcp-server/dist/index.js"], "env": { "SIXTYDB_API_KEY": "sk_your_api_key_here", "SIXTYDB_API_BASE_URL": "https://api.60db.ai" } } } } ``` ### 3. Use with Claude Restart Claude Desktop and start using 60db voice AI: ``` You: Can you convert "Hello, world!" to speech using a female English voice? Claude: I'll use the 60db TTS tool to generate speech for you. ``` ## Available Tools | Category | Tools | | ---------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | **Voices** | List voices, get voice details, create custom voices | | **TTS** | Synthesize speech, view generation history | | **STT** | Transcribe audio, view transcription history | | **Workspaces** | Manage team workspaces and members | | **Memory & RAG** | Ingest memories, upload documents (91+ formats with OCR), hybrid semantic search, context assembly, manage collections, track spend (10 tools total) | | **Authz** | Fetch effective permissions map, probe single `(resource, action)` checks — useful before attempting gated actions | | **Analytics** | Track usage statistics | | **Billing** | View plans, subscriptions, and invoices | ## Features ### Text-to-Speech Generate natural speech in 40+ languages: ```typescript theme={null} // Synthesize speech { "text": "Hello, world!", "voice_id": "voice_abc123", "speed": 1.0, "output_format": "mp3" } ``` ### Speech-to-Text Transcribe audio with speaker identification: ```typescript theme={null} // Transcribe audio { "audio_url": "https://example.com/meeting.mp3", "language": "en-US", "diarization": true, "timestamps": true } ``` ### Voice Cloning Create custom voices from audio samples: ```typescript theme={null} // Create custom voice { "name": "Brand Voice", "sample_url": "https://example.com/sample.mp3", "language": "en-US" } ``` ### Memory & RAG Give your AI assistant a persistent, searchable long-term memory: ```typescript theme={null} // Store a user preference { "tool": "sixtydb_memory_ingest", "arguments": { "text": "User prefers dark mode and metric units", "type": "user" } } // Upload a document (PDF, DOCX, scanned image, etc.) { "tool": "sixtydb_memory_upload_document", "arguments": { "file_path": "/Users/me/Documents/handbook.pdf", "collection": "company_handbook", "type": "knowledge" } } // Search with hybrid semantic + keyword recall { "tool": "sixtydb_memory_search", "arguments": { "query": "What is the refund policy?", "alpha": 0.8, "mode": "fast" } } // Assemble LLM-ready context for RAG { "tool": "sixtydb_memory_context", "arguments": { "query": "Tell me about my customer's recent orders" } } ``` All memory operations are [pay-as-you-go](/api-reference/memory/pricing) — the tool response surfaces `charged`, `balance`, and `tx` details in the billing footer on every call. ### Authorization checks Before attempting a gated action, query the Cerbos permission map so the agent can fail fast on disallowed operations: ```typescript theme={null} // Fetch the full permission map (once per session) { "tool": "sixtydb_get_permissions", "arguments": {} } // Probe a specific action { "tool": "sixtydb_check_permission", "arguments": { "resource": "memory", "action": "create" } } ``` This lets the agent hide or disable features that the user's role can't use, instead of watching for `403 POLICY_DENY` responses mid-flow. ## Architecture ``` ┌─────────────────────────────────────────────────────────────┐ │ Claude (AI Assistant) │ └─────────────────────────┬───────────────────────────────────┘ │ │ MCP Protocol (stdio) │ ┌─────────────────────────▼───────────────────────────────────┐ │ 60db MCP Server │ ├─────────────────────────────────────────────────────────────┤ │ Voice: sixtydb_tts_synthesize, sixtydb_stt_transcribe │ │ sixtydb_list_voices, sixtydb_create_voice │ │ Memory: sixtydb_memory_ingest, sixtydb_memory_search │ │ sixtydb_memory_upload_document │ │ sixtydb_memory_context, sixtydb_memory_get_usage │ │ Authz: sixtydb_get_permissions, sixtydb_check_permission│ │ Admin: sixtydb_list_workspaces, sixtydb_get_usage_stats │ │ sixtydb_list_invoices │ │ ... 30+ tools total │ └─────────────────────────┬───────────────────────────────────┘ │ │ HTTPS (JWT or API key) │ ┌─────────────────────────▼───────────────────────────────────┐ │ 60db API │ │ (api.60db.ai) │ │ │ │ Every request passes through: │ │ Cerbos policy check → Memory billing → Controller │ └─────────────────────────────────────────────────────────────┘ ``` ## Next Steps * [Installation](/mcp-server/installation) - Install and configure the server * [Configuration](/mcp-server/configuration) - Set up authentication * [Quick Start](/mcp-server/quickstart) - Run your first request * [Claude Desktop](/mcp-server/claude-desktop) - Integration guide ## Requirements * **Node.js**: 18 or higher * **API Key**: 60db API key from dashboard * **MCP Client**: Claude Desktop or custom MCP client ## Support * **Documentation**: [https://docs.60db.ai](https://docs.60db.ai) * **GitHub**: [https://github.com/60db-ai/mcp](https://github.com/60db-ai/mcp) * **Issues**: [https://github.com/60db-ai/mcp/issues](https://github.com/60db-ai/mcp/issues)