Context continuity MCP server for developers. Solves AI conversation loss - automatically preserves architecture and code discussions for clear productivity advantage. Works seamlessly with Cursor and VS Code. Next: Full system design and architecture assistance.
CodeArchitect provides 4 tools with explicit registrations in src/server.ts. Tool naming follows verb_noun convention (store_session, get_session, search_session, codearchitect_help). Descriptions are present and moderately detailed (ranging 194-500+ characters), exceeding the 10-character minimum but many fall short of the optimal 50-200 character range for LLM parsing. Input schemas are visible and properly typed with JSON Schema. However, output schemas are NOT documented, the response structure is entirely undocumented, which forces LLMs to guess what fields are returned. Parameters lack specific format constraints (e.g., date format YYYY-MM-DD is mentioned in descriptions but not enforced in schema). Error handling is mentioned conceptually but not shown in the actual tool implementations. Notably, store_session's description is extremely long (1500+ characters) and includes operational workflow details that should be in separate guidance, not the tool definition, this violates the pattern of keeping tool descriptions concise and focused. Security posture is weak: the store_session tool writes to user home directories (~/.codearchitect/sessions/) without explicit permission gates, and stores arbitrary conversation data without sanitization or audit logging.
Get help about CodeArchitect MCP features. Use when user says "use codearchitect" without specifying a feature, or asks "what can codearchitect do?". Returns available features with examples and next steps.
[CodeArchitect] Retrieve saved sessions from knowledge base === WHEN TO USE === - User says: "use codearchitect get_session", "retrieve [topic]", "show me my notes on...", "get session" - User wants to reference past conversations - Before re-explaining something === RETRIEVAL BEHAVIOR === 1. Retrieves from main folder: ~/.codearchitect/sessions/ 2. Lists all sessions if no params provided 3. Filters by filename or date if provided 4. Returns structured session content with metadata === PARAMETERS === - filename (optional): Specific session filename/topic to retrieve (case-insensitive search) - date (optional): Date folder to search in (YYYY-MM-DD format) - sessionsDir (optional): Custom sessions directory - format (optional): 'json' | 'toon' | 'auto' - Output format for LLM optimization. Default: 'auto' - limit (optional): Limit number of sessions returned (for list). Default: all === USAGE PATTERNS === Pattern 1: List all sessions - User says: "get_session" or "show all my sessions" - Tool returns: List of all sessions with topics, dates, file sizes Pattern 2: Get specific session by topic - User says: "get_session authentication" or "get_session 'auth implementation'" - Tool returns: Full session content for matching topic Pattern 3: Get sessions by date - User says: "get_session 2025-11-19" or "sessions from 2025-11-19" - Tool returns: All sessions from that date Pattern 4: Get session with specific format - User says: "get_session authentication format: json" or "get_session authentication format: toon" - Tool returns: Session in specified format (for LLM optimization) === RESPONSE === If getting specific session: - Returns session content with metadata (filename, topic, date, file path) - If messages parsed: returns structured message array - Format: 'json' or 'toon' based on what was used If listing sessions: - Returns array of session metadata (filename, topic, date, size) - Sorted by date (newest first) - Shows count of sessions found
No output schemas documented for any tool. LLMs cannot infer response structure, forcing them to guess field names and types. This violates pattern:response-shaper and pattern:tool-chain requirements.
store_session description is 1500+ characters, far exceeding the 1024-character maximum. Operational workflow details (export file location, step-by-step instructions) belong in separate documentation, not the tool definition. This wastes tokens and buries the core 'save conversation' intent.
Date parameters (store_session.exportFilename, get_session.date, search_session.date/dateFrom/dateTo) lack explicit format constraints in JSON Schema. Descriptions mention YYYY-MM-DD format, but schema does not enforce it. LLMs cannot reliably parse or validate date inputs without formal constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |
[CodeArchitect] Search across all stored sessions === WHEN TO USE === - User says: "search my sessions for...", "find discussions about...", "search for..." - User is looking for specific topics or keywords - User remembers discussing something but not the exact session name - User wants to explore the knowledge base === SEARCH BEHAVIOR === 1. Searches through all sessions in ~/.codearchitect/sessions/ 2. Returns matching sessions sorted by relevance score 3. Includes context snippets showing where the query matched 4. Supports date filtering and result limiting === PARAMETERS === - query (required): Search query (keywords, topic, phrase) - date (optional): Filter by specific date (YYYY-MM-DD format) - dateFrom (optional): Filter from date (YYYY-MM-DD format) - dateTo (optional): Filter to date (YYYY-MM-DD format) - limit (optional): Limit number of results returned. Default: all - sessionsDir (optional): Custom sessions directory === SEARCH FEATURES === - Case-insensitive matching - Searches in topic, content, and message bodies - Relevance scoring (weighted: topic > content > messages) - Context snippets (150 chars before/after match) - Multiple match counting - Date range filtering === USAGE PATTERNS === Pattern 1: Simple keyword search - User says: "search_session authentication" - Tool searches for "authentication" in all sessions - Returns sessions with relevance scores and snippets Pattern 2: Phrase search - User says: "search_session 'database design'" - Tool searches for exact phrase - Returns matching sessions with context Pattern 3: Search with date filtering - User says: "search_session 'API design' date: '2025-11-19'" - Tool searches only in sessions from that date - Returns results sorted by relevance Pattern 4: Search with date range - User says: "search_session 'architecture' dateFrom: '2025-11-01' dateTo: '2025-11-30'" - Tool searches in date range - Returns results sorted by relevance Pattern 5: Limit results - User says: "search_session 'implementation' limit: 10" - Tool returns top 10 most relevant results === RESPONSE === Returns: - List of matching sessions with metadata - Relevance score for each (0-1) - Number of matches found in each session - Context snippets showing query in context - Total count of matching sessions - Original query for reference
[CodeArchitect] Save important conversations for future reference. === WHEN TO USE === - User says: "use codearchitect store_session", "save this conversation", "store this", "remember this" - User wants to keep/remember an important discussion - After solving complex problems or making architecture decisions === STORAGE BEHAVIOR === 1. ALWAYS saves to main folder: ~/.codearchitect/sessions/ (user's home directory) 2. OPTIONALLY saves to project folder if user specifies projectDir parameter 3. NEVER auto-detects project - only use projectDir if user explicitly provides it === CONVERSATION SOURCE (CHOOSE ONE) === OPTION 1: Export File (RECOMMENDED - Most Reliable) - User exports conversation from Cursor/VS Code to .codearchitect/exports/ folder - Tool automatically detects and processes the export file - No need to manually extract conversation WORKFLOW: 1. User exports chat: In Cursor/VS Code → three dots (⋯) → "Export Chat" 2. User saves to: ~/.codearchitect/exports/ folder (main location - always the same) 3. User says: "use codearchitect store_session" 4. Tool automatically finds newest export file and stores it If multiple exports exist: - Tool processes newest file (modified in last 10 minutes) - If user wants specific file: use exportFilename parameter (e.g., "resolve_mcp") OPTION 2: Direct Conversation Parameter (Fallback) - Only use if export file method doesn't work - Extract FULL conversation from your context window - Format as plain text with USER/ASSISTANT markers OR JSON array === PARAMETERS === - conversation (optional): Direct conversation text/JSON - only use if export file unavailable - exportFilename (optional): Pattern to match specific export file (e.g., "resolve_mcp" matches "cursor_resolve_mcp_configuration_issues.md") - topic (optional): Session topic - auto-extracted if not provided - projectDir (optional): Project directory path - if provided, saves to both main and project folders - format (optional): "plain" or "messages" - default "plain" === WORKFLOW (EXPORT METHOD) === Step 1: User exports conversation to ~/.codearchitect/exports/ folder (main location) Step 2: User says "use codearchitect store_session" Step 3: Tool detects export file automatically from main exports folder Step 4: Tool parses and stores conversation Step 5: Tool saves to main sessions folder (always) + optionally project folder if projectDir specified === ERROR HANDLING === If no export file found: - Return clear instructions with OS/IDE-specific export folder path - Tell user exactly where to save the export file - Provide step-by-step instructions === RESPONSE === After saving, return concise message: - "Detected export file '[filename]', saved to main: [topic-name]. Next: use codearchitect get_session [topic-name]" - Or if projectDir provided: "Detected export file '[filename]', saved to main: [topic-name] and project: [topic-name]. Next: use codearchitect get_session [topic-name]"
store_session parameter 'conversation' description includes example JSON format inline ('e.g. ["role": "user/assistant", "content": "..."]'). LLMs tend to copy example values and formats literally, increasing risk of malformed input. Use formal JSON Schema constraints instead of text examples.
codearchitect_help parameter 'feature' lacks enum constraint. Should be enum: ['store_session', 'get_session', 'search_session'] to prevent LLMs from inventing feature names.
No error handling documented. Tools do not show how they handle missing files, invalid dates, permission errors, or disk full scenarios. Error descriptions are absent, LLMs cannot recover from failures.
store_session is a WRITE tool (creates/modifies files) but provides no dry-run, confirmation, or undo capability. Agents cannot safely test the operation before executing. Pattern:confirmation-request is violated.
No audit logging or permission gates. store_session writes to user home directory (~/.codearchitect/sessions/) without verifying the caller has authority or logging who called it. Sensitive conversation data is stored without encryption or access control indicators.
search_session.limit parameter is type 'number' but lacks min/max bounds in schema. LLMs could pass negative, zero, or millions, causing performance issues or crashes. Should enforce 1 - 1000 range.
get_session parameter 'format' includes enum value 'toon' (Token-Oriented Object Notation from @toon-format/toon package), but LLMs do not know what this format is without external docs. Consider renaming to 'compact' or including a parameter description explaining the token-reduction benefit.