AGI MCP Server

Instructions

Implementation Reference

• src/memory-manager.js:907-927 (handler)

  Core handler function that executes the tool logic: queries the database for memory clusters similar to the given clusterId using vector embedding cosine similarity (pgvector <=> operator).

  async findSimilarClusters(clusterId, threshold = 0.7) {
    try {
      const embeddingVector = `[${Array(1536).fill(0).join(',')}]`;
      
      const results = await this.db.execute(sql`
        SELECT 
          mc2.*,
          1 - (mc1.centroid_embedding <=> mc2.centroid_embedding) as similarity
        FROM memory_clusters mc1
        CROSS JOIN memory_clusters mc2
        WHERE mc1.id = ${clusterId}
          AND mc2.id != ${clusterId}
          AND 1 - (mc1.centroid_embedding <=> mc2.centroid_embedding) >= ${threshold}
        ORDER BY similarity DESC
      `);
      
      return results.rows || [];
    } catch (error) {
      console.warn('Similar clusters query failed:', error.message);
      return [];
    }

• mcp.js:387-404 (registration)

  Registration of the tool in the MCP server's tools array, defining name, description, and input schema.

  {
    name: "find_similar_clusters",
    description: "Find clusters similar to a given cluster",
    inputSchema: {
      type: "object",
      properties: {
        cluster_id: {
          type: "string",
          description: "UUID of the reference cluster"
        },
        threshold: {
          type: "number",
          description: "Minimum similarity threshold",
          default: 0.7
        }
      },
      required: ["cluster_id"]
    }

• mcp.js:649-654 (handler)

  MCP server request handler that dispatches to memoryManager.findSimilarClusters and returns JSON-formatted response.

  case "find_similar_clusters":
    const similarClusters = await memoryManager.findSimilarClusters(
      args.cluster_id,
      args.threshold || 0.7
    );
    return { content: [{ type: "text", text: JSON.stringify(similarClusters, null, 2) }] };