M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

// ============================================================================ // Unified Graph Tools - Consolidated for Better UX // 6 tools: memory_node, memory_edge, memory_batch, memory_lock, get_task_context, memory_clear // Reduced from 22 tools while maintaining all functionality // ============================================================================ import type { Tool } from "@modelcontextprotocol/sdk/types.js"; export const GRAPH_TOOLS: Tool[] = [ // ============================================================================ // TOOL 1: memory_node - All node operations // ============================================================================ { name: "memory_node", description: `Manage memory nodes (knowledge entries). Operations: add, get, update, delete, query, search. Nodes store conversation details, decisions, file references, concepts. All nodes automatically get semantic embeddings for later retrieval via vector_search_nodes. Use IDs to reference nodes instead of repeating details. Examples: - Add: memory_node(operation='add', type='memory', properties={title: 'X', content: 'Y'}) - Get: memory_node(operation='get', id='memory-123') - Query: memory_node(operation='query', type='todo', filters={status: 'pending'}) - Search: memory_node(operation='search', query='authentication code') [semantic search by default, automatic fallback to keyword search if embeddings disabled or no results found]`, inputSchema: { type: "object", properties: { operation: { type: "string", enum: ["add", "get", "update", "delete", "query", "search"], description: "Operation to perform on nodes" }, id: { type: "string", description: "Node ID (required for get, update, delete)" }, type: { type: "string", enum: ["todo", "todoList", "memory", "file", "function", "class", "module", "concept", "person", "project", "custom"], description: "Node type (required for add, optional for query)" }, properties: { type: "object", description: "Node properties for add/update operations. Nested objects are automatically flattened (e.g., {details: {files: ['a.ts']}} becomes {details_files: ['a.ts']}). For best performance, flatten complex structures client-side before calling.", additionalProperties: true }, filters: { type: "object", description: "Property filters for query (e.g., {status: 'pending', priority: 'high'})", additionalProperties: true }, query: { type: "string", description: "Search query text for search operation (semantic search with automatic fallback to keyword search)" }, options: { type: "object", description: "Search options: {limit: 100, offset: 0, types: ['todo', 'memory']}", additionalProperties: true }, confirm: { type: "boolean", description: "Set to true to confirm a destructive operation (used with confirmationId for delete operations)" }, confirmationId: { type: "string", description: "Confirmation token from a previous preview request (required when confirm=true for delete)" } }, required: ["operation"] } }, // ============================================================================ // TOOL 2: memory_edge - All edge/relationship operations // ============================================================================ { name: "memory_edge", description: `Manage relationships between nodes. Operations: add, delete, get, neighbors, subgraph. Build knowledge graphs by linking nodes (e.g., 'file depends_on module', 'todo part_of project'). Examples: - Add: memory_edge(operation='add', source='todo-1', target='project-2', type='part_of') - Get edges: memory_edge(operation='get', node_id='todo-1', direction='both') - Neighbors: memory_edge(operation='neighbors', node_id='todo-1', edge_type='depends_on') - Subgraph: memory_edge(operation='subgraph', node_id='project-1', depth=2)`, inputSchema: { type: "object", properties: { operation: { type: "string", enum: ["add", "delete", "get", "neighbors", "subgraph"], description: "Operation to perform on edges/relationships" }, source: { type: "string", description: "Source node ID (required for add)" }, target: { type: "string", description: "Target node ID (required for add)" }, edge_id: { type: "string", description: "Edge ID (required for delete)" }, node_id: { type: "string", description: "Node ID (required for get, neighbors, subgraph)" }, type: { type: "string", enum: ["contains", "depends_on", "relates_to", "implements", "calls", "imports", "assigned_to", "parent_of", "blocks", "references"], description: "Edge type (required for add)" }, edge_type: { type: "string", description: "Filter by edge type (optional for neighbors)" }, direction: { type: "string", enum: ["in", "out", "both"], description: "Edge direction for get operation (default: both)" }, depth: { type: "number", description: "Traversal depth for neighbors/subgraph (default: 1 for neighbors, 2 for subgraph)" }, properties: { type: "object", description: "Edge properties for add operation. Nested objects are automatically flattened to conform to Neo4j constraints.", additionalProperties: true } }, required: ["operation"] } }, // ============================================================================ // TOOL 3: memory_batch - Bulk operations // ============================================================================ { name: "memory_batch", description: `Perform bulk operations on multiple nodes/edges efficiently. Operations: add_nodes, update_nodes, delete_nodes, add_edges, delete_edges. Use for batch processing (e.g., creating multiple todos, bulk updates). All properties are automatically flattened if nested. Examples: - Add nodes: memory_batch(operation='add_nodes', nodes=[{type: 'todo', properties: {...}}, ...]) - Update nodes: memory_batch(operation='update_nodes', updates=[{id: 'todo-1', properties: {status: 'completed'}}, ...]) - Delete nodes: memory_batch(operation='delete_nodes', ids=['todo-1', 'todo-2'])`, inputSchema: { type: "object", properties: { operation: { type: "string", enum: ["add_nodes", "update_nodes", "delete_nodes", "add_edges", "delete_edges"], description: "Batch operation to perform" }, nodes: { type: "array", description: "Array of nodes for add_nodes: [{type: 'todo', properties: {...}}, ...]. Properties are automatically flattened if nested.", items: { type: "object", properties: { type: { type: "string" }, properties: { type: "object", additionalProperties: true } } } }, updates: { type: "array", description: "Array of updates for update_nodes: [{id: 'todo-1', properties: {...}}, ...]. Properties are automatically flattened if nested.", items: { type: "object", properties: { id: { type: "string" }, properties: { type: "object", additionalProperties: true } } } }, ids: { type: "array", description: "Array of IDs for delete_nodes/delete_edges", items: { type: "string" } }, edges: { type: "array", description: "Array of edges for add_edges: [{source: 'a', target: 'b', type: 'depends_on', properties: {...}}, ...]. Properties are automatically flattened if nested.", items: { type: "object", properties: { source: { type: "string" }, target: { type: "string" }, type: { type: "string" }, properties: { type: "object", additionalProperties: true } } } }, confirm: { type: "boolean", description: "Set to true to confirm a destructive operation (used with confirmationId for delete_nodes/delete_edges)" }, confirmationId: { type: "string", description: "Confirmation token from a previous preview request (required when confirm=true for batch deletes)" } }, required: ["operation"] } }, // ============================================================================ // TOOL 4: memory_lock - Multi-agent locking // ============================================================================ { name: "memory_lock", description: `Manage locks for multi-agent coordination. Operations: acquire, release, query_available, cleanup. Prevent race conditions when multiple agents work on same tasks. Examples: - Acquire: memory_lock(operation='acquire', node_id='todo-1', agent_id='worker-1') - Release: memory_lock(operation='release', node_id='todo-1', agent_id='worker-1') - Query: memory_lock(operation='query_available', type='todo', filters={status: 'pending'}) - Cleanup: memory_lock(operation='cleanup')`, inputSchema: { type: "object", properties: { operation: { type: "string", enum: ["acquire", "release", "query_available", "cleanup"], description: "Lock operation to perform" }, node_id: { type: "string", description: "Node ID (required for acquire, release)" }, agent_id: { type: "string", description: "Agent ID (required for acquire, release)" }, timeout_ms: { type: "number", description: "Lock timeout in milliseconds (default: 300000 = 5 min)" }, type: { type: "string", description: "Node type filter for query_available" }, filters: { type: "object", description: "Property filters for query_available", additionalProperties: true } }, required: ["operation"] } }, // ============================================================================ // TOOL 5: get_task_context - Context isolation (specialized) // ============================================================================ { name: "get_task_context", description: "Get filtered task context based on agent type (PM/worker/QC). Server-side context isolation for multi-agent workflows. PM agents get full context (100%), workers get minimal context (<10% - only files, dependencies, requirements), QC agents get requirements + worker output for verification. Implements 90%+ context reduction for worker agents.", inputSchema: { type: "object", properties: { taskId: { type: "string", description: "Task node ID to retrieve context for" }, agentType: { type: "string", enum: ["pm", "worker", "qc"], description: "Agent type requesting context - determines filtering level" } }, required: ["taskId", "agentType"] } }, // ============================================================================ // TOOL 6: memory_clear - Dangerous operation (deserves own tool) // ============================================================================ { name: "memory_clear", description: "Clear data from the graph. SAFETY: To clear all data, you MUST explicitly pass type='ALL'. To clear specific node types, pass the node type. Returns counts of deleted nodes and edges. REQUIRES CONFIRMATION: First call without 'confirm' to get preview and confirmationId, then call again with confirm=true and the confirmationId to execute.", inputSchema: { type: "object", properties: { type: { type: "string", enum: ["ALL", "todo", "todoList", "memory", "file", "function", "class", "module", "concept", "person", "project", "custom"], description: "Node type to clear, or 'ALL' to clear entire graph (use with extreme caution!). Required parameter." }, confirm: { type: "boolean", description: "Set to true to confirm the clear operation (used with confirmationId)" }, confirmationId: { type: "string", description: "Confirmation token from the preview request (required when confirm=true)" } }, required: ["type"] } } ];