Skip to main content
Glama
rspraneeth

retrieval-only-RAG

by rspraneeth
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

retrieval-only-RAG

A PDF retrieval tool wrapped as an MCP server. It handles the R in RAG — your IDE agent (Cursor, Kiro, Claude Code) handles generation.

PDFs ──> [ load → chunk → embed → store → retrieve ]
                          │
                 returns matching chunks
                          │
              [ MCP server wraps the retriever ]
                          │
     IDE agent calls it ──┘  →  IDE agent writes the answer

No LLM inside this tool. Embeddings run locally (no cloud key needed).

Setup

python -m venv .venv
.venv\Scripts\activate        # Windows
pip install -r requirements.txt

Related MCP server: MyDocsMCP

Usage

Index your PDFs — drop PDF files into pdfs/ then run:

python -m pdf_rag.cli index

Only new or changed PDFs are processed on subsequent runs — unchanged files are skipped. Deleted PDFs have their chunks removed automatically.

Search — retrieve the top-k chunks for a question:

python -m pdf_rag.cli search "What is the difference between ArrayList and LinkedList?"

Output includes source filename, page number, and similarity score for each chunk.

MCP Server

Exposes one tool — search_pdfs(query) — that any MCP-compatible IDE agent can call.

python mcp_server.py

Claude Code (.mcp.json in project root)

A .mcp.json is already included in this repo:

{
  "mcpServers": {
    "pdf-rag": {
      "command": "C:\\Projects\\Retrieval\\.venv\\Scripts\\python.exe",
      "args": ["C:\\Projects\\Retrieval\\mcp_server.py"],
      "cwd": "C:\\Projects\\Retrieval"
    }
  }
}

Update the paths to match your machine, then Claude Code picks it up automatically.

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "pdf-rag": {
      "command": "path/to/.venv/Scripts/python.exe",
      "args": ["path/to/mcp_server.py"],
      "cwd": "path/to/project"
    }
  }
}

Once connected, ask your IDE agent a question about your PDFs — it calls search_pdfs, gets the chunks, and writes the answer. You own retrieval; the agent owns generation.

Configuration (config.yaml)

pdf_folder: pdfs              # folder to scan for PDFs
vector_store: vector_store    # where ChromaDB persists the index
embedding_model: BAAI/bge-small-en-v1.5   # local HuggingFace model
top_k: 5                      # chunks returned per query

Project structure

pdf_rag/
  config.py      # load + validate config.yaml
  indexer.py     # PDF loading, chunking, embedding, ChromaDB persistence
  retriever.py   # similarity search + result formatting
  cli.py         # index / search commands
mcp_server.py    # MCP wrapper exposing search_pdfs()
config.yaml
requirements.txt
.mcp.json        # Claude Code MCP config (update paths for your machine)
pdfs/            # drop your PDFs here (not committed)
vector_store/    # ChromaDB index + manifest.json (not committed)

How the RAG split works

Layer

Who does it

How

Retrieval

This tool

LlamaIndex + ChromaDB + local embeddings

Augmentation

MCP protocol

Retrieved chunks injected into agent context

Generation

IDE agent

Cursor / Kiro / Claude Code answers from chunks

The MCP server is editor-agnostic — swap Cursor for Kiro (or any MCP client) by changing only the connection config, no code changes needed.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Resources

Related MCP Connectors

Related MCP Servers

  • mcp-context

    ericlimabr
    F
    license
    Not graded
    quality
    D
    maintenance
    Local MCP server that provides semantic search (RAG) over code repositories, enabling AI clients like Claude and Gemini to access project context without manual re-upload.
    -

  • pdf-context

    mematcha
    A
    license
    A
    quality
    D
    maintenance
    A local-first MCP server that ingests PDFs, extracts structure, and provides semantic search and sequential navigation tools for AI clients to query and learn from documents.
    10
    MIT