Omi Memories MCP Server

Implementation Reference

• src/server.ts:46-86 (handler)

  The handler function that executes the 'fetch-memories' tool logic. It makes an API request to fetch conversations for a specific user ID, formats the data into a readable text summary including IDs, times, titles, overviews, categories, emojis, and transcripts, and returns it as MCP content. Includes error handling for Axios errors.
  try { const response = await axios.get(`${EXPRESS_API_URL}/memories`, { params: { uid: SPECIFIC_USER_ID } }); const conversations: Conversation[] = response.data.conversations; // Format conversations for better readability const formattedText = conversations.map(conv => { return ` Memory: ${conv.id} Time: ${new Date(conv.created_at).toLocaleString()} Title: ${conv.structured.title} Overview: ${conv.structured.overview} Category: ${conv.structured.category} Emoji: ${conv.structured.emoji} Transcript: ${conv.transcript_segments.map(segment => ` ${segment.speaker}: ${segment.text}`).join('\n')} ------------------- `.trim(); }).join('\n\n'); return { content: [{ type: "text", text: formattedText }] }; } catch (error) { if (axios.isAxiosError(error)) { return { content: [{ type: "text", text: `Error: Failed to fetch memories: ${error.message}` }], isError: true }; } throw error; } }
• src/server.ts:43-87 (registration)

  Registration of the 'fetch-memories' tool using McpServer.tool() method, specifying the tool name, an empty input schema (no parameters), and an inline async handler function.
  "fetch-memories", {}, // No parameters needed since we're using a specific user ID async () => { try { const response = await axios.get(`${EXPRESS_API_URL}/memories`, { params: { uid: SPECIFIC_USER_ID } }); const conversations: Conversation[] = response.data.conversations; // Format conversations for better readability const formattedText = conversations.map(conv => { return ` Memory: ${conv.id} Time: ${new Date(conv.created_at).toLocaleString()} Title: ${conv.structured.title} Overview: ${conv.structured.overview} Category: ${conv.structured.category} Emoji: ${conv.structured.emoji} Transcript: ${conv.transcript_segments.map(segment => ` ${segment.speaker}: ${segment.text}`).join('\n')} ------------------- `.trim(); }).join('\n\n'); return { content: [{ type: "text", text: formattedText }] }; } catch (error) { if (axios.isAxiosError(error)) { return { content: [{ type: "text", text: `Error: Failed to fetch memories: ${error.message}` }], isError: true }; } throw error; } } );
• src/server.ts:44-44 (schema)

  Empty Zod schema object indicating the tool takes no input parameters.
  {}, // No parameters needed since we're using a specific user ID
• src/server.ts:16-39 (helper)

  TypeScript interface defining the Conversation type used to type the API response data in the handler, including structured metadata and transcript segments.
  interface Conversation { id: string; created_at: string; started_at: string; finished_at: string; source: string; structured: { emoji: string; events: any[]; overview: string; title: string; action_items: any[]; category: string; }; transcript_segments: { is_user: boolean; start: number; end: number; text: string; speaker_id: number; speaker: string; person_id: null; }[]; }