Local RAG MCP Server by shinpr: Setup for Claude, Cursor and Other Apps

• Searches your indexed documents

  Searches everything you have indexed by meaning and by keywords, and returns the best matching passages with their file and position. You can limit the number of results.

  Try: “Find what our docs say about ERR_CONNECTION_REFUSED.”

  Files · query_documents

• Reads the text around a search result

  Returns the passages just before and after a search hit from the same document, so the assistant can read it in context. Two before and two after by default.

  Try: “Show me the paragraphs around that result.”

  Files · read_chunk_neighbors

• Adds one file to the index

  Reads a PDF, Word, text or Markdown file by its full path, splits it into passages and stores them in the local index. Adding the same path again replaces the old entry.

  Try: “Index /Users/me/docs/handbook.pdf.”

  Write · ingest_file

• Adds text or a web page you already have

  Indexes text, Markdown or HTML that the assistant passes in, under a source name you choose. The server does not fetch web pages itself.

  Try: “Index this page content under the name pricing-page.”

  Write · ingest_data

• Updates the index to match your folders

  Starts a background job that adds new and changed files, skips unchanged ones, and removes index entries for files that no longer exist on disk.

  Try: “Sync my docs folder with the index.”

  Delete · sync_start

• Checks on a running sync

  Reports the progress of a sync job started earlier.

  Try: “Is the sync finished?”

  Read · sync_status

• Shows which files are indexed

  Lists the supported files under your document folders and whether each one has been indexed.

  Try: “Which of my documents are not indexed yet?”

  Files · list_files

• Removes a file from the index

  Deletes the stored passages of one indexed file or one text item from the local index, and reports how many passages it removed.

  Try: “Remove handbook.pdf from the index.”

  Delete · delete_file

• Shows index statistics

  Returns how many documents and passages are indexed, memory use, uptime and the search mode.

  Try: “How big is my index?”

  Read · status

Watch out for

• It writes an index to disk and can remove entries

  ingest and sync write the search index to the DB_PATH folder, and sync_start and delete_file remove entries from it. They do not delete your source files, so set DB_PATH to a folder you control. github.com

• It reads every file in your document folders

  Everything under BASE_DIR becomes searchable by the assistant. Point it at a folder with only the documents you want to share with your AI app. github.com

• An optional reranker command runs on your machine

  If you set RAG_RERANK_CMD, the server runs that command for each search. Leave it unset unless you wrote the command yourself. github.com

• PDF figure captions download extra models

  If the assistant ingests a PDF with the visual option, the server downloads a captioning model on first use: about 250 MB for the fast profile and about 1.7 GB for quality. Plain text ingestion does not need it. github.com

Security advisories