Letta MCP Server

/** * Enhanced tool descriptions with detailed examples and best practices */ export const enhancedDescriptions = { // Agent Management Tools create_agent: { description: 'Create a new Letta agent with specified configuration. After creation, use attach_tool to add capabilities, attach_memory_block to configure memory, or prompt_agent to start conversations.', longDescription: `Creates a new Letta agent with customizable LLM and embedding models. Best practices: - Always set up core memory blocks (persona, human) after creation - Attach relevant tools based on the agent's purpose - Test the agent with a simple prompt to verify functionality Common workflows: 1. Create agent → Set up memory → Attach tools → Test 2. Use letta_agent_wizard prompt for guided creation`, examples: [ { scenario: 'Create a research assistant', params: { name: 'research-assistant', description: 'An agent specialized in research and information gathering', model: 'openai/gpt-4', }, }, { scenario: 'Create a code helper with specific model', params: { name: 'code-helper', description: 'Assists with programming tasks and code review', model: 'anthropic/claude-3-opus', embedding: 'openai/text-embedding-3-large', }, }, ], }, prompt_agent: { description: 'Send a message to an agent and get a response. Ensure the agent has necessary tools attached (see attach_tool) first. Use list_agents to find agent IDs.', longDescription: `Sends a message to a Letta agent and retrieves its response. The agent will use its attached tools and memory to provide contextual responses. Prerequisites: - Agent must exist and be properly configured - Memory blocks should be attached for context - Relevant tools should be attached for functionality Response includes: - Agent's message content - Tool usage information - Memory updates if applicable`, examples: [ { scenario: 'Simple conversation', params: { agent_id: 'agent-123', message: 'Hello, can you help me understand your capabilities?', }, }, { scenario: 'Task-oriented prompt', params: { agent_id: 'research-agent-456', message: 'Please research the latest developments in quantum computing and summarize the key findings.', }, }, ], }, attach_tool: { description: 'Attach one or more tools (by ID or name) to an agent. If a name corresponds to an MCP tool not yet in Letta, it will be registered first. Find tools with list_mcp_tools_by_server or create custom ones with upload_tool. Use list_agent_tools to verify attachment.', longDescription: `Attaches tools to an agent, expanding its capabilities. Supports both existing Letta tools and MCP tools that can be auto-registered. Tool attachment strategies: - Use tool_ids for known Letta tools - Use tool_names for MCP tools (auto-registers if needed) - Batch attach multiple tools for efficiency After attachment: - Verify with list_agent_tools - Test tool functionality with prompt_agent - Monitor performance impact`, examples: [ { scenario: 'Attach multiple tools by ID', params: { agent_id: 'agent-123', tool_ids: ['tool-456', 'tool-789'], }, }, { scenario: 'Attach MCP tools by name', params: { agent_id: 'agent-123', tool_names: ['web_search', 'file_reader', 'calculator'], }, }, ], }, // Memory Management Tools create_memory_block: { description: 'Create a new memory block in the Letta system. Common labels: "persona", "human", "system". Use attach_memory_block to link to agents, or update_memory_block to modify later.', longDescription: `Creates a memory block that can be attached to agents for persistent context. Memory block types: - persona: Agent's identity and behavior - human: Information about the user - system: System-level instructions - custom: Domain-specific memory Best practices: - Keep persona concise and focused - Update human block with user preferences - Use system blocks for operational guidelines`, examples: [ { scenario: 'Create persona memory', params: { name: 'research_assistant_persona', label: 'persona', value: 'You are a meticulous research assistant who values accuracy and comprehensive analysis. You cite sources and acknowledge uncertainty.', }, }, { scenario: 'Create user context memory', params: { name: 'user_preferences', label: 'human', value: 'The user prefers technical explanations and appreciates detailed breakdowns of complex topics.', agent_id: 'agent-123', }, }, ], }, list_passages: { description: "Retrieve the memories in an agent's archival memory store (paginated query). Use create_passage to add new memories, modify_passage to edit, or delete_passage to remove them.", longDescription: `Retrieves archival memories (passages) for an agent with pagination and search capabilities. Use cases: - Audit agent's knowledge base - Search for specific information - Manage memory capacity - Export agent knowledge Search tips: - Use keywords for targeted results - Combine with pagination for large datasets - Sort by date for recent memories`, examples: [ { scenario: 'Get all passages', params: { agent_id: 'agent-123', limit: 50, }, }, { scenario: 'Search specific topic', params: { agent_id: 'agent-123', search: 'machine learning algorithms', limit: 20, }, }, ], }, // Tool Management list_mcp_tools_by_server: { description: 'List all available tools for a specific MCP server. Use list_mcp_servers first to see available servers, then add_mcp_tool_to_letta to import tools into Letta.', longDescription: `Discovers available tools from a connected MCP server for potential integration with Letta agents. Workflow: 1. List MCP servers to find available sources 2. List tools from desired server 3. Review tool capabilities and requirements 4. Import selected tools with add_mcp_tool_to_letta 5. Attach to agents as needed Filter options help find specific tool types or capabilities.`, examples: [ { scenario: 'List all tools from a server', params: { mcp_server_name: 'web-tools-server', }, }, { scenario: 'Search for specific tools', params: { mcp_server_name: 'utility-server', filter: 'calculator', pageSize: 20, }, }, ], }, // Bulk Operations bulk_attach_tool_to_agents: { description: 'Attaches a specified tool to multiple agents based on filter criteria (name or tags). Use list_agents to find agents and list_mcp_tools_by_server or upload_tool to get tool IDs.', longDescription: `Efficiently attaches a tool to multiple agents matching specified criteria. Use cases: - Deploy new capability across agent fleet - Standardize tool sets for agent types - Update agents after tool improvements Filtering options: - By name pattern (substring match) - By tags (exact or partial match) - Combine filters for precision`, examples: [ { scenario: 'Attach to all research agents', params: { tool_id: 'web-search-tool-123', agent_name_filter: 'research', }, }, { scenario: 'Attach to tagged agents', params: { tool_id: 'calculator-tool-456', agent_tag_filter: 'mathematical,scientific', }, }, ], }, // Import/Export export_agent: { description: "Export an agent's configuration to a JSON file and optionally upload it. Use import_agent to recreate the agent later, or clone_agent for a quick copy. Use list_agents to find agent IDs.", longDescription: `Exports complete agent configuration including tools, memory, and settings for backup or migration. Export includes: - Agent metadata and configuration - Attached tools and their settings - Memory blocks and content - Custom configurations Options: - Save locally or upload to cloud - Base64 encoding for API transport - Automatic file naming`, examples: [ { scenario: 'Export for backup', params: { agent_id: 'agent-123', output_path: 'backups/agent-123-backup.json', }, }, { scenario: 'Export and upload', params: { agent_id: 'agent-123', upload_to_xbackbone: true, return_base64: true, }, }, ], }, // Debugging get_agent_summary: { description: "Provides a concise summary of an agent's configuration, including core memory snippets and attached tool/source names. Use list_agents to find agent IDs. Follow up with modify_agent to change settings or attach_tool to add capabilities.", longDescription: `Retrieves a comprehensive summary of an agent's current state for quick assessment. Summary includes: - Basic configuration (name, model, etc.) - Memory block excerpts - Tool inventory with names - Recent activity indicators - Performance metrics Use for: - Quick status checks - Debugging conversations - Pre-modification review - Documentation`, examples: [ { scenario: 'Get full agent overview', params: { agent_id: 'agent-123', }, }, ], }, }; /** * Get enhanced description for a tool * @param {string} toolName - Name of the tool * @returns {Object} Enhanced description object */ export function getEnhancedDescription(toolName) { return ( enhancedDescriptions[toolName] || { description: `Tool ${toolName} - No enhanced description available`, longDescription: 'Please refer to the basic tool description.', examples: [], } ); } /** * Generate markdown documentation for a tool * @param {string} toolName - Name of the tool * @returns {string} Markdown formatted documentation */ export function generateToolDocumentation(toolName) { const enhanced = getEnhancedDescription(toolName); let doc = `# ${toolName}\n\n`; doc += `${enhanced.description}\n\n`; if (enhanced.longDescription) { doc += `## Detailed Description\n\n${enhanced.longDescription}\n\n`; } if (enhanced.examples && enhanced.examples.length > 0) { doc += `## Examples\n\n`; enhanced.examples.forEach((example, index) => { doc += `### Example ${index + 1}: ${example.scenario}\n\n`; doc += '```json\n'; doc += JSON.stringify(example.params, null, 2); doc += '\n```\n\n'; }); } return doc; }