search
Find information in your knowledge base by searching across stored Markdown files. This tool helps you locate specific content within your personal semantic graph.
Instructions
Search for content across the knowledge base
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes |
Implementation Reference
- The handler function for the MCP tool named 'search'. This is a ChatGPT/OpenAI compatible adapter that performs a search using the underlying search_notes tool, formats the results in the expected OpenAI schema, and returns them as a content list. Includes registration via @mcp.tool() decorator.@mcp.tool(description="Search for content across the knowledge base") async def search( query: str, context: Context | None = None, ) -> List[Dict[str, Any]]: """ChatGPT/OpenAI MCP search adapter returning a single text content item. Args: query: Search query (full-text syntax supported by `search_notes`) context: Optional FastMCP context passed through for auth/session data Returns: List with one dict: `{ "type": "text", "text": "{...JSON...}" }` where the JSON body contains `results`, `total_count`, and echo of `query`. """ track_mcp_tool("search") logger.info(f"ChatGPT search request: query='{query}'") try: # ChatGPT tools don't expose project parameter, so use default project config = ConfigManager().config default_project = config.default_project # Call underlying search_notes with sensible defaults for ChatGPT results = await search_notes.fn( query=query, project=default_project, # Use default project for ChatGPT page=1, page_size=10, # Reasonable default for ChatGPT consumption search_type="text", # Default to full-text search context=context, ) # Handle string error responses from search_notes if isinstance(results, str): logger.warning(f"Search failed with error: {results[:100]}...") search_results = { "results": [], "error": "Search failed", "error_details": results[:500], # Truncate long error messages } else: # Format successful results for ChatGPT formatted_results = _format_search_results_for_chatgpt(results) search_results = { "results": formatted_results, "total_count": len(results.results), # Use actual count from results "query": query, } logger.info(f"Search completed: {len(formatted_results)} results returned") # Return in MCP content array format as required by OpenAI return [{"type": "text", "text": json.dumps(search_results, ensure_ascii=False)}] except Exception as e: logger.error(f"ChatGPT search failed for query '{query}': {e}") error_results = { "results": [], "error": "Internal search error", "error_message": str(e)[:200], } return [{"type": "text", "text": json.dumps(error_results, ensure_ascii=False)}]
- Helper function that formats SearchResponse objects into the list of dictionaries expected by ChatGPT/OpenAI MCP clients (with id, title, url fields).def _format_search_results_for_chatgpt(results: SearchResponse) -> List[Dict[str, Any]]: """Format search results according to ChatGPT's expected schema. Returns a list of result objects with id, title, and url fields. """ formatted_results = [] for result in results.results: formatted_result = { "id": result.permalink or f"doc-{len(formatted_results)}", "title": result.title if result.title and result.title.strip() else "Untitled", "url": result.permalink or "", } formatted_results.append(formatted_result) return formatted_results
- Pydantic schemas for search queries and responses (SearchQuery, SearchResult, SearchResponse), used by the underlying search_notes tool invoked by the 'search' MCP tool.class SearchQuery(BaseModel): """Search query parameters. Use ONE of these primary search modes: - permalink: Exact permalink match - permalink_match: Path pattern with * - text: Full-text search of title/content (supports boolean operators: AND, OR, NOT) Optionally filter results by: - types: Limit to specific item types - entity_types: Limit to specific entity types - after_date: Only items after date Boolean search examples: - "python AND flask" - Find items with both terms - "python OR django" - Find items with either term - "python NOT django" - Find items with python but not django - "(python OR flask) AND web" - Use parentheses for grouping """ # Primary search modes (use ONE of these) permalink: Optional[str] = None # Exact permalink match permalink_match: Optional[str] = None # Glob permalink match text: Optional[str] = None # Full-text search (now supports boolean operators) title: Optional[str] = None # title only search # Optional filters types: Optional[List[str]] = None # Filter by type entity_types: Optional[List[SearchItemType]] = None # Filter by entity type after_date: Optional[Union[datetime, str]] = None # Time-based filter @field_validator("after_date") @classmethod def validate_date(cls, v: Optional[Union[datetime, str]]) -> Optional[str]: """Convert datetime to ISO format if needed.""" if isinstance(v, datetime): return v.isoformat() return v def no_criteria(self) -> bool: return ( self.permalink is None and self.permalink_match is None and self.title is None and self.text is None and self.after_date is None and self.types is None and self.entity_types is None ) def has_boolean_operators(self) -> bool: """Check if the text query contains boolean operators (AND, OR, NOT).""" if not self.text: # pragma: no cover return False # Check for common boolean operators with correct word boundaries # to avoid matching substrings like "GRAND" containing "AND" boolean_patterns = [" AND ", " OR ", " NOT ", "(", ")"] text = f" {self.text} " # Add spaces to ensure we match word boundaries return any(pattern in text for pattern in boolean_patterns) class SearchResult(BaseModel): """Search result with score and metadata.""" title: str type: SearchItemType score: float entity: Optional[Permalink] = None permalink: Optional[str] content: Optional[str] = None file_path: str metadata: Optional[dict] = None # IDs for v2 API consistency entity_id: Optional[int] = None # Entity ID (always present for entities) observation_id: Optional[int] = None # Observation ID (for observation results) relation_id: Optional[int] = None # Relation ID (for relation results) # Type-specific fields category: Optional[str] = None # For observations from_entity: Optional[Permalink] = None # For relations to_entity: Optional[Permalink] = None # For relations relation_type: Optional[str] = None # For relations class SearchResponse(BaseModel): """Wrapper for search results.""" results: List[SearchResult] current_page: int
- The underlying 'search_notes' tool that performs the actual search API call and error handling, invoked by the 'search' ChatGPT adapter.@mcp.tool( description="Search across all content in the knowledge base with advanced syntax support.", ) async def search_notes( query: str, project: Optional[str] = None, page: int = 1, page_size: int = 10, search_type: str = "text", types: List[str] | None = None, entity_types: List[str] | None = None, after_date: Optional[str] = None, context: Context | None = None, ) -> SearchResponse | str: """Search across all content in the knowledge base with comprehensive syntax support. This tool searches the knowledge base using full-text search, pattern matching, or exact permalink lookup. It supports filtering by content type, entity type, and date, with advanced boolean and phrase search capabilities. Project Resolution: Server resolves projects in this order: Single Project Mode → project parameter → default project. If project unknown, use list_memory_projects() or recent_activity() first. ## Search Syntax Examples ### Basic Searches - `search_notes("my-project", "keyword")` - Find any content containing "keyword" - `search_notes("work-docs", "'exact phrase'")` - Search for exact phrase match ### Advanced Boolean Searches - `search_notes("my-project", "term1 term2")` - Find content with both terms (implicit AND) - `search_notes("my-project", "term1 AND term2")` - Explicit AND search (both terms required) - `search_notes("my-project", "term1 OR term2")` - Either term can be present - `search_notes("my-project", "term1 NOT term2")` - Include term1 but exclude term2 - `search_notes("my-project", "(project OR planning) AND notes")` - Grouped boolean logic ### Content-Specific Searches - `search_notes("research", "tag:example")` - Search within specific tags (if supported by content) - `search_notes("work-project", "category:observation")` - Filter by observation categories - `search_notes("team-docs", "author:username")` - Find content by author (if metadata available) ### Search Type Examples - `search_notes("my-project", "Meeting", search_type="title")` - Search only in titles - `search_notes("work-docs", "docs/meeting-*", search_type="permalink")` - Pattern match permalinks - `search_notes("research", "keyword", search_type="text")` - Full-text search (default) ### Filtering Options - `search_notes("my-project", "query", types=["entity"])` - Search only entities - `search_notes("work-docs", "query", types=["note", "person"])` - Multiple content types - `search_notes("research", "query", entity_types=["observation"])` - Filter by entity type - `search_notes("team-docs", "query", after_date="2024-01-01")` - Recent content only - `search_notes("my-project", "query", after_date="1 week")` - Relative date filtering ### Advanced Pattern Examples - `search_notes("work-project", "project AND (meeting OR discussion)")` - Complex boolean logic - `search_notes("research", "\"exact phrase\" AND keyword")` - Combine phrase and keyword search - `search_notes("dev-notes", "bug NOT fixed")` - Exclude resolved issues - `search_notes("archive", "docs/2024-*", search_type="permalink")` - Year-based permalink search Args: query: The search query string (supports boolean operators, phrases, patterns) project: Project name to search in. Optional - server will resolve using hierarchy. If unknown, use list_memory_projects() to discover available projects. page: The page number of results to return (default 1) page_size: The number of results to return per page (default 10) search_type: Type of search to perform, one of: "text", "title", "permalink" (default: "text") types: Optional list of note types to search (e.g., ["note", "person"]) entity_types: Optional list of entity types to filter by (e.g., ["entity", "observation"]) after_date: Optional date filter for recent content (e.g., "1 week", "2d", "2024-01-01") context: Optional FastMCP context for performance caching. Returns: SearchResponse with results and pagination info, or helpful error guidance if search fails Examples: # Basic text search results = await search_notes("project planning") # Boolean AND search (both terms must be present) results = await search_notes("project AND planning") # Boolean OR search (either term can be present) results = await search_notes("project OR meeting") # Boolean NOT search (exclude terms) results = await search_notes("project NOT meeting") # Boolean search with grouping results = await search_notes("(project OR planning) AND notes") # Exact phrase search results = await search_notes("\"weekly standup meeting\"") # Search with type filter results = await search_notes( "meeting notes", types=["entity"], ) # Search with entity type filter results = await search_notes( "meeting notes", entity_types=["observation"], ) # Search for recent content results = await search_notes( "bug report", after_date="1 week" ) # Pattern matching on permalinks results = await search_notes( "docs/meeting-*", search_type="permalink" ) # Title-only search results = await search_notes( "Machine Learning", search_type="title" ) # Complex search with multiple filters results = await search_notes( "(bug OR issue) AND NOT resolved", types=["entity"], after_date="2024-01-01" ) # Explicit project specification results = await search_notes("project planning", project="my-project") """ track_mcp_tool("search_notes") # Avoid mutable-default-argument footguns. Treat None as "no filter". types = types or [] entity_types = entity_types or [] # Create a SearchQuery object based on the parameters search_query = SearchQuery() # Set the appropriate search field based on search_type if search_type == "text": search_query.text = query elif search_type == "title": search_query.title = query elif search_type == "permalink" and "*" in query: search_query.permalink_match = query elif search_type == "permalink": search_query.permalink = query else: search_query.text = query # Default to text search # Add optional filters if provided (empty lists are treated as no filter) if entity_types: search_query.entity_types = [SearchItemType(t) for t in entity_types] if types: search_query.types = types if after_date: search_query.after_date = after_date async with get_client() as client: active_project = await get_active_project(client, project, context) logger.info(f"Searching for {search_query} in project {active_project.name}") try: response = await call_post( client, f"/v2/projects/{active_project.external_id}/search/", json=search_query.model_dump(), params={"page": page, "page_size": page_size}, ) result = SearchResponse.model_validate(response.json()) # Check if we got no results and provide helpful guidance if not result.results: logger.info( f"Search returned no results for query: {query} in project {active_project.name}" ) # Don't treat this as an error, but the user might want guidance # We return the empty result as normal - the user can decide if they need help return result except Exception as e: logger.error(f"Search failed for query '{query}': {e}, project: {active_project.name}") # Return formatted error message as string for better user experience return _format_search_error_response(active_project.name, str(e), query, search_type)