Which integrations are available for this server?

Used for logging and monitoring in the future AWS migration. Used for document storage (PDFs) in the future AWS migration. Used for orchestration or deployment of services in the future AWS migration. Used as the vector store in the future AWS migration via Amazon OpenSearch Serverless.

How do I use Insurance Claims MCP RAG Agent?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Insurance Claims MCP RAG Agent What is the coverage limit for water damage in policy X?" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

Insurance Claims MCP RAG Agent

by SwapnilMundhekar

Overview Schema Related Servers Score Discussions

Python

Local

MCP-Based Insurance Claims RAG Agent

Project Overview

This project builds an enterprise-style AI agent for insurance claims and policy intelligence.

The system is designed to answer questions over insurance policy and regulatory documents, retrieve relevant clauses, and expose retrieval capabilities through MCP tools. The local version runs with Ollama and a local vector database. The future cloud version will migrate to Claude on Amazon Bedrock and AWS-native retrieval infrastructure.

The goal is not to create a simple chatbot. The goal is to build a governed AI system where the language model proposes actions, the orchestrator validates them, MCP tools expose controlled capabilities, and retrieval outputs remain traceable to source documents.

Related MCP server: AskDocs MCP Server

Core Principle

The model proposes.
The orchestrator validates.
The MCP server executes.
The retrieval layer provides grounded evidence.

Current Status

Build Part	Module	Status
1	Project skeleton and local environment	Complete
2	PDF ingestion	Next
3	Cleaning and metadata extraction	Planned
4	Section-aware semantic chunking	Planned
5	Embeddings and local vector store	Planned
6	Retrieval testing	Planned
7	Ollama local LLM integration	Planned
8	MCP policy knowledge server	Planned
9	Governed tool-calling orchestrator	Planned
10	Streamlit app, tracing, documentation, AWS plan	Planned

System Architecture

flowchart TD
    A[User Question] --> B[Local Agent Application]
    B --> C[Ollama Local LLM]
    C --> D{Is a tool required?}

    D -->|No| E[Generate Grounded Answer]
    D -->|Yes| F[Tool Calling Orchestrator]

    F --> G[Validate Tool Name]
    G --> H[Validate Tool Arguments]
    H --> I[Apply Policy and Safety Checks]
    I --> J[MCP Client]

    J --> K[MCP Policy Knowledge Server]
    K --> L[Retrieval Service]
    L --> M[Local Vector Store]
    M --> N[Semantic Chunks]
    N --> O[Insurance PDF Documents]

    K --> P[Structured Tool Result]
    P --> C
    C --> Q[Final Answer with Citations]

    F --> R[Trace Logs]
    F --> S[Audit Events]

Local-to-Cloud Migration View

flowchart LR
    A[Local PDF Folder] --> B[AWS S3 Document Bucket]
    C[Ollama Local LLM] --> D[Claude on Amazon Bedrock]
    E[ChromaDB Local Vector Store] --> F[Amazon OpenSearch Serverless]
    G[Local MCP Server] --> H[AWS Hosted MCP Services]
    I[Local Logs] --> J[Amazon CloudWatch]
    K[Local Streamlit App] --> L[Cloud Frontend or Internal App]

Tool Calling Lifecycle

sequenceDiagram
    participant User
    participant LLM as Local LLM
    participant Orchestrator
    participant MCP as MCP Server
    participant Store as Vector Store

    User->>LLM: Ask policy or claim question
    LLM->>Orchestrator: Propose structured tool call
    Orchestrator->>Orchestrator: Validate tool name
    Orchestrator->>Orchestrator: Validate arguments
    Orchestrator->>Orchestrator: Apply safety checks
    Orchestrator->>MCP: Execute approved MCP tool
    MCP->>Store: Retrieve relevant chunks
    Store-->>MCP: Return matched chunks
    MCP-->>Orchestrator: Return structured tool result
    Orchestrator-->>LLM: Provide observation
    LLM-->>User: Generate grounded answer with citations

Retrieval Pipeline

flowchart TD
    A[Insurance PDF Documents] --> B[PDF Page Extraction]
    B --> C[Text Cleaning]
    C --> D[Document Metadata Assignment]
    D --> E[Heading and Clause Detection]
    E --> F[Section Aware Chunking]
    F --> G[Semantic Boundary Detection]
    G --> H[Token Limit Guardrail]
    H --> I[Embedding Generation]
    I --> J[Vector Store Index]
    J --> K[Top K Retrieval]
    K --> L[Answer Context with Citations]

Planned Knowledge Base

The local policy documents are stored outside the Git repository.

C:\Users\SSS\Desktop\AI Project\pdf policy documents

The documents are expected to cover these categories:

Category	Example Documents	Purpose
Policy documents	NRMA Car Insurance PDS, NRMA Home Insurance PDS	Coverage, exclusions, claim conditions
Claims handling	AFCA claims handling guidance	Fair claims handling and dispute context
Regulatory guidance	ASIC INFO 253, ASIC RG 271	Claims handling and complaint obligations
Industry conduct	General Insurance Code of Practice	Customer treatment and industry standards
Operational risk	APRA CPS 230	Enterprise risk, resilience, and operational controls

PDF files are not committed to GitHub.

Planned MCP Tools

Tool Name	Purpose	Risk Level
`search_policy_documents`	Search policy and regulatory document chunks	Low
`get_policy_clause`	Retrieve a specific policy section or clause	Low
`compare_policy_coverage`	Compare claim facts against retrieved clauses	Medium
`get_complaint_obligations`	Retrieve complaint handling obligations	Low
`draft_customer_response`	Draft customer-facing claim communication	Medium
`log_audit_event`	Record tool call and workflow trace	Low

Side-effecting tools such as sending email, updating claim status, or escalating a claim will require explicit approval gates in later versions.

Technology Stack

Layer	Local MVP	Future AWS Version
LLM	Ollama local model	Claude on Amazon Bedrock
Embeddings	SentenceTransformers or local embedding model	Amazon Bedrock embeddings
Vector store	ChromaDB	Amazon OpenSearch Serverless or Bedrock Knowledge Bases
Tool interface	MCP Python server	AWS-hosted MCP services
Orchestration	Custom Python orchestrator	Containerized service or AWS Lambda/ECS
UI	Streamlit	Streamlit, React, or internal enterprise frontend
Logs	Local log files	Amazon CloudWatch
Document storage	Local PDF folder	Amazon S3

Repository Structure

insurance-claims-mcp-rag/
|
|-- app/
|-- config/
|-- data/
|   |-- raw/
|   |-- processed/
|   |-- chunks/
|
|-- docs/
|-- logs/
|-- scripts/
|-- src/
|   |-- ingestion/
|   |-- chunking/
|   |-- retrieval/
|   |-- llm/
|   |-- mcp_server/
|   |-- orchestrator/
|
|-- tests/
|-- vector_store/
|-- .env.example
|-- .gitignore
|-- README.md
|-- requirements.txt

Build Roadmap

gantt
    title Project Build Roadmap
    dateFormat  YYYY-MM-DD
    axisFormat  %d %b

    section Foundation
    Project skeleton                  :done, p1, 2026-06-24, 1d
    PDF ingestion                     :active, p2, 2026-06-25, 1d
    Cleaning and metadata             :p3, 2026-06-26, 1d

    section Retrieval
    Section aware semantic chunking    :p4, 2026-06-27, 1d
    Embeddings and vector store        :p5, 2026-06-28, 1d
    Retrieval testing                  :p6, 2026-06-29, 1d

    section Agent Layer
    Ollama LLM integration             :p7, 2026-06-30, 1d
    MCP policy server                  :p8, 2026-07-01, 1d
    Tool calling orchestrator          :p9, 2026-07-02, 1d

    section Productization
    Streamlit app and documentation    :p10, 2026-07-03, 1d

Local Setup

1. Clone the repository

git clone https://github.com/SwapnilMundhekar/insurance-claims-mcp-rag.git
cd insurance-claims-mcp-rag

2. Create virtual environment

python -m venv .venv

3. Activate virtual environment on Windows

.\.venv\Scripts\Activate.ps1

4. Install dependencies

pip install -r requirements.txt

5. Create `.env`

Copy .env.example to .env and update the local PDF path.

PDF_SOURCE_DIR=C:\Users\SSS\Desktop\AI Project\pdf policy documents

Environment Variables

PROJECT_NAME=insurance-claims-mcp-rag
PROJECT_ROOT=YOUR_PROJECT_PATH
PDF_SOURCE_DIR=YOUR_POLICY_PDF_FOLDER_PATH

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=qwen2.5:7b-instruct

VECTOR_DB_DIR=.\vector_store\chroma
CHUNK_OUTPUT_DIR=.\data\chunks
PROCESSED_OUTPUT_DIR=.\data\processed

What Part 1 Contains

Part 1 includes:

Area	Implemented
Project folder structure	Yes
Python virtual environment plan	Yes
Requirements file	Yes
Git ignore rules	Yes
Environment example	Yes
Source module boundaries	Yes
PDF data excluded from GitHub	Yes

Next Build Step

Part 2 will implement PDF ingestion.

Expected output:

data/processed/extracted_pages.json

The ingestion output will contain:

{
  "document_name": "example-policy-document.pdf",
  "page_number": 1,
  "text": "Extracted page text..."
}

Future Demo Assets

After the Streamlit application is built, a real demo animation will be added here:

docs/assets/demo.gif

The README will then include:

![Application Demo](docs/assets/demo.gif)

No fake demo screenshots or generated performance metrics are included in this repository.

Engineering Concepts Demonstrated

This project is designed to demonstrate:

Concept	How it appears in the project
RAG	Retrieval over policy and regulatory documents
Semantic chunking	Section-aware document splitting
MCP	Tool interface for controlled retrieval actions
Tool calling	LLM proposes structured tool use
Orchestration	Python validates and executes tool requests
Observability	Trace logs and audit events
Governance	Approval gates for risky actions
Cloud migration	Local Ollama to Claude on AWS Bedrock

Design Constraints

The system will follow these constraints:

Do not commit PDFs to GitHub.
Do not hardcode secrets.
Do not allow the LLM to directly execute actions.
Validate every tool call before execution.
Preserve document metadata for citations.
Return uncertainty when retrieved context is insufficient.
Keep local development cloud-portable.

Author

Swapnil Mundhekar

GitHub: SwapnilMundhekar

License

This project is currently for personal learning, portfolio development, and interview preparation.

A formal license can be added later.

This server cannot be installed

license - not found

quality - not tested

maintenance

How are these scores calculated?

Maintenance

–Maintainers

–Response time

–Release cycle

–Releases (12mo)

Commit activity

Resources

GitHub Repository

Need Help?

Related Servers

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Latest Blog Posts

Lightport: Open-Sourcing Glama's AI Gateway
By punkpeye on April 27, 2026.
open source
OpenAI
Tool Definition Quality Score (TDQS)
By punkpeye on April 3, 2026.
mcp
The Hackers Who Tracked My Sleep Cycle
By punkpeye on March 26, 2026.
security

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/SwapnilMundhekar/insurance-claims-mcp-rag'

If you have feedback or need assistance with the MCP directory API, please join our Discord server