All-Search MCP Server

You're implementing a REST API for a RAG system. You are to use FastAPI and LangChain. Below are instructions on all of the different endpoints you need to implement. # API Endpoint Definitions ## Collections Manage vector store collections. - POST /collections - Creates a new collection. - Request Body: JSON containing collection details (e.g., {'name': 'my_collection'}). - Response: Details of the created collection or confirmation. - GET /collections - Lists all available collections. - Response: List of collection identifiers or objects. - GET /collections/{collection_id} - Retrieves details of a specific collection. - Path Parameter: collection_id - The ID of the collection to retrieve. - Response: Details of the specified collection. - PUT /collections/{collection_id} - Updates/replaces an existing collection (e.g., rename). - Path Parameter: collection_id - The ID of the collection to update. - Request Body: JSON containing the full updated collection details. - Response: Details of the updated collection. - PATCH /collections/{collection_id} - Partially updates an existing collection. - Path Parameter: collection_id - The ID of the collection to update. - Request Body: JSON containing the specific fields to update. - Response: Details of the updated collection. - DELETE /collections/{collection_id} - Deletes a specific collection. - Path Parameter: collection_id - The ID of the collection to delete. - Response: Confirmation of deletion. ## Documents (within Collections) Manage documents within a specific collection (RAG functionality). - POST /collections/{collection_id}/documents - Indexes (adds) a new document to the specified collection. - Path Parameter: collection_id - The ID of the collection to add the document to. - Request Body: The document data to be indexed. - Response: Identifier or details of the indexed document. - GET /collections/{collection_id}/documents - Lists all documents within a specific collection. - Path Parameter: collection_id - The ID of the collection. - Query Parameters (Optional): - query={search_terms}: Filter documents based on search terms. - limit={N}: Limit the number of results. - offset={M}: Skip the first M results (for pagination). - Response: List of document identifiers or objects within the collection. - GET /collections/{collection_id}/documents/{document_id} - Retrieves a specific document from a collection. - Path Parameters: - collection_id: The ID of the collection. - document_id: The ID of the document to retrieve. - Response: The content or details of the specified document. - PUT /collections/{collection_id}/documents/{document_id} - Updates/replaces an existing document in a collection. - Path Parameters: - collection_id: The ID of the collection. - document_id: The ID of the document to update. - Request Body: The full updated document data. - Response: Details of the updated document. - PATCH /collections/{collection_id}/documents/{document_id} - Partially updates an existing document in a collection. - Path Parameters: - collection_id: The ID of the collection. - document_id: The ID of the document to update. - Request Body: JSON containing the specific fields/parts of the document to update. - Response: Details of the updated document. - DELETE /collections/{collection_id}/documents/{document_id} - Deletes a specific document from a collection. - Path Parameters: - collection_id: The ID of the collection. - document_id: The ID of the document to delete. - Response: Confirmation of deletion. - POST /collections/{collection_id}/documents/search (Alternative Search) - Performs a search within a specific collection using potentially complex criteria. - Use this if GET with query parameters is insufficient (e.g., requires a request body). - Path Parameter: collection_id - The ID of the collection to search within. - Request Body: JSON containing search criteria. - Response: List of matching documents. ## LangChain Integration Please setup this application with LangChain document loaders, text splitters and vector stores. ### Document Loaders You should use the `UploadFile` type from FastAPI for the inputs to the API for uploading documents. Then, use the `Blob` class from `langchain_core.documents` to load the uploaded file as a blob. Finally, use the `MimeTypeBasedParser` from `langchain_community.document_loaders.parsers.generic` to parse the blob into a document. Here is some example code, and the types of documents you should support: ```python from langchain_community.document_loaders.parsers import BS4HTMLParser, PDFMinerParser from langchain_community.document_loaders.parsers.generic import MimeTypeBasedParser from langchain_community.document_loaders.parsers.msword import MsWordParser from langchain_community.document_loaders.parsers.txt import TextParser HANDLERS = { "application/pdf": PDFMinerParser(), "text/plain": TextParser(), "text/html": BS4HTMLParser(), "application/msword": MsWordParser(), "application/vnd.openxmlformats-officedocument.wordprocessingml.document": ( MsWordParser() ), } SUPPORTED_MIMETYPES = sorted(HANDLERS.keys()) MIMETYPE_BASED_PARSER = MimeTypeBasedParser( handlers=HANDLERS, fallback_parser=None, ) ``` ### Text Splitters For text splitting, you should use the `RecursiveCharacterTextSplitter` from `langchain_text_splitters`. Set the following parameters: `chunk_size=1000, chunk_overlap=200`. ### Vector Stores For the vector store, use the PGVector LangChain integration. For connection details, use environment variables. Import from the `langchain_postgres` package. You should also use postgres to create collections, and fetch/search/delete/create collections.