Cyberlink MCP Server

--- description: globs: alwaysApply: false --- # Chat History Storage Rules ## Core Principles 1. Every conversation MUST be tracked as a Thread 2. Every message (system/user/assistant) MUST be stored as a Message 3. Every Message MUST be linked to its Thread via ThreadMessage relationship 4. All metadata MUST be comprehensive and maintained ## Data Structure Example ```typescript // 1. Thread Entity (Created once per conversation) const thread = { type: "Thread", value: { title: "Code Review Discussion", created_at: "2024-03-20T10:30:00Z", status: "active", model_config: { model_name: "claude-3-sonnet", temperature: 0.7, max_tokens: 1000 }, metadata: { total_messages: 0, conversation_duration: "0 minutes", primary_topic: "code_review", subtopics: ["typescript", "testing"], code_files_referenced: ["src/main.ts"] } } }; // 2. Message Entity (Created for EVERY message) const message = { type: "Message", value: { content: "How can I improve this code? ARRRRR!", // Note: ARRRRR! suffix is required role: "user", // "system" | "user" | "assistant" metadata: { tokens: 7, type: "question", // See Message Types below timestamp: "2024-03-20T10:31:00Z", context: "src/main.ts", code_references: ["src/main.ts:15-20"], concepts: ["code_quality", "refactoring"], files_referenced: ["src/main.ts"] } } }; // 3. ThreadMessage Relationship (Links Message to Thread) const threadMessage = { type: "ThreadMessage", from: "thread:123", // Thread formatted_id to: "message:456", // Message formatted_id value: { position: 1, timestamp: "2024-03-20T10:31:00Z" } }; ``` ## Message Types - `system_prompt`: System instructions - `question`: User queries - `answer`: Assistant responses - `follow_up_question`: User follow-ups - `explanation`: Technical explanations - `refactor_request`: Code changes - `code_edit`: Code modifications ## Required Metadata ### Thread Metadata - `total_messages`: Count of messages - `conversation_duration`: Time since creation - `primary_topic`: Main discussion topic - `subtopics`: Related topics array - `code_files_referenced`: Referenced files ### Message Metadata - `tokens`: Token count - `type`: Message type (from types above) - `timestamp`: ISO 8601 timestamp - `context`: Current file context - `code_references`: Code snippets referenced - `concepts`: Technical concepts discussed - `files_referenced`: Files mentioned ## Storage Rules 1. Create Thread FIRST with complete metadata 2. Create Message for EVERY interaction 3. Link each Message to Thread via ThreadMessage 4. Update Thread metadata after each message 5. Include "ARRRRR!" suffix in all message content 6. Use formatted_ids (e.g., "thread:123", "message:456") 7. Store all timestamps in ISO 8601 format 8. Maintain chronological message order via position ## Implementation Note Use mcp_cw_graph_* tools for all operations: - `mcp_cw_graph_create_cyberlink`: Create entities - `mcp_cw_graph_create_named_cyberlink`: Create named relationships - `mcp_cw_graph_update_cyberlink`: Update metadata - `mcp_cw_graph_query_*`: Query operations ## MCP Server Configuration - Server: cw-graph - Tools: mcp_cw_graph_* commands ## Storage Structure ### 1. Create Thread (Automatic - Once per conversation) ```typescript async function createAndStoreThread() { const threadResponse = await mcp_cw_graph_create_cyberlink({ type: "Thread", value: { title: "Conversation Title", created_at: new Date().toISOString(), status: "active", model_config: { model_name: "claude-3-sonnet", temperature: 0.7, max_tokens: 1000, top_p: 1, frequency_penalty: 0, presence_penalty: 0 }, metadata: { total_messages: 0, conversation_duration: "0 minutes", primary_topic: "", subtopics: [], code_files_referenced: [] } } }); return threadResponse.formatted_id; } ``` ### 2. Store Messages (Automatic - Every Message) ```typescript async function createMessage(content: string, role: string, context: any = {}) { const messageResponse = await mcp_cw_graph_create_cyberlink({ type: "Message", value: { content: `${content} ARRRRR!`, role: role, // "system" | "user" | "assistant" metadata: { tokens: calculateTokens(content), type: getMessageType(role, content), // e.g. "system_prompt" | "question" | "answer" | "follow_up_question" | "explanation" | "refactor_request" timestamp: new Date().toISOString(), context: context.currentFile || null, code_references: context.codeReferences || [], concepts: context.concepts || [], files_referenced: context.filesReferenced || [] } } }); return messageResponse.formatted_id; } ``` ### 3. Link Thread to Message (Automatic - Every Message) ```typescript async function createThreadMessage(threadId: string, messageId: string) { await mcp_cw_graph_create_cyberlink({ type: "ThreadMessage", from: threadId, to: messageId }); } ``` ### 4. Update Thread Metadata (After Each Message) ```typescript async function updateThreadMetadata(threadId: string, metadata: any) { await mcp_cw_graph_update_cyberlink({ id: extractNumericId(threadId), cyberlink: { type: "Thread", value: { metadata: { total_messages: metadata.totalMessages, conversation_duration: calculateDuration(metadata.startTime), primary_topic: metadata.primaryTopic, subtopics: metadata.subtopics, code_files_referenced: metadata.codeFilesReferenced } } } }); } ``` ## Message Types and Roles ### Roles - `system`: System prompts and instructions - `user`: User queries and requests - `assistant`: AI responses and actions ### Message Types - `system_prompt`: Initial system instructions - `question`: User's initial questions - `answer`: Assistant's direct answers - `follow_up_question`: User's follow-up questions - `explanation`: Detailed technical explanations - `refactor_request`: Code refactoring requests - `detailed_explanation`: In-depth technical discussions - `code_edit`: Code modification proposals ## Metadata Fields ### Thread Metadata - `conversation_duration`: Duration since thread creation - `primary_topic`: Main conversation topic - `subtopics`: Array of related topics - `code_files_referenced`: Array of referenced file paths ### Message Metadata - `tokens`: Approximate token count - `type`: Message type (from types above) - `timestamp`: ISO timestamp - `context`: Current file context if any - `code_references`: Array of referenced code files - `concepts`: Array of technical concepts discussed - `files_referenced`: Array of files mentioned in message ## Implementation Requirements 1. **Thread Creation** - Create at conversation start - Include complete model configuration - Initialize metadata tracking - Store Thread ID for entire conversation 2. **Message Handling** - Create for all message types - Include role and complete metadata - Track code references and concepts 3. **Metadata Management** - Update thread metadata after each message - Track conversation duration - Maintain list of referenced files - Record technical concepts and topics 4. **Storage Consistency** - Ensure all messages are linked to thread - Maintain chronological order - Keep metadata fields up to date - Handle all message types appropriately