Knowledge Base (RAG) Integration

Mockarty supports connecting external knowledge bases to enrich AI agent responses with your company-specific documentation, API specifications, coding standards, and other reference materials.

About URLs in examples: All examples use localhost:5770 as the default Mockarty address. If your instance runs on a remote server, replace localhost:5770 with its actual address (e.g. https://mockarty.company.com or http://192.168.1.50:5770). See Tips & Useful Features for details.

What is RAG?

Simple explanation: Imagine you hire a new developer. They are smart, but they do not know your company’s APIs, coding standards, or internal tools. RAG is like giving them a searchable handbook. Before answering a question, they check the handbook first, so their answers are accurate and specific to your company.

RAG (Retrieval-Augmented Generation) is a pattern where the AI searches a knowledge base for relevant information before generating a response. Instead of relying solely on the model’s training data, RAG retrieves your specific documents and injects them into the conversation context.

This means your AI agents can:

• Know your company’s API specifications and generate accurate mocks
• Follow your team’s report format and coding standards
• Reference internal documentation when answering questions
• Provide context-aware suggestions based on your project’s history

How It Works in Mockarty

Knowledge bases integrate through the MCP Marketplace as a RAG source type. Once connected, the AI agent gets two tools:

When a user asks the AI agent a question, the agent can automatically search the knowledge base for relevant context before responding.

Provider Comparison

Before diving into details, here is a quick comparison to help you choose:

Recommendation for beginners: Start with AnythingLLM – it is a single Docker container with a web UI and drag-and-drop document upload. You can switch to a more advanced provider later without changing your Mockarty configuration.

Supported RAG Providers

RAGFlow

Best for: Deep document understanding (tables, images, complex layouts)

• Apache 2.0 license
• Excellent PDF/DOCX parsing (DeepDoc engine)
• Built-in MCP server support
• REST API on port 9380

Quick Start:

git clone https://github.com/infiniflow/ragflow.git
cd ragflow
docker compose -f docker/docker-compose-base.yml up -d

Default URL: http://localhost:9380

After starting:

1. Open RAGFlow UI at http://localhost:9380
2. Create a dataset and upload documents
3. Get your API key from Settings

AnythingLLM

Best for: Simplest setup (single container, drag & drop)

• MIT license
• Built-in vector database (LanceDB)
• Single Docker container
• Web UI with drag & drop upload

Quick Start:

docker run -d -p 3001:3001 \
  --name anythingllm \
  -v anythingllm_data:/app/server/storage \
  mintplexlabs/anythingllm

Default URL: http://localhost:3001

After starting:

1. Open AnythingLLM at http://localhost:3001
2. Complete the setup wizard
3. Create a workspace and upload documents
4. Get your API key from Settings > Developer

Dify

Best for: Advanced RAG workflows with visual pipeline builder

• Apache 2.0 license
• Visual workflow editor
• 30+ vector database options
• Bidirectional MCP support

Quick Start:

git clone https://github.com/langgenius/dify.git
cd dify/docker
docker compose up -d

Default URL: http://localhost:3000

After starting:

1. Open Dify at http://localhost:3000
2. Create a Knowledge Base and upload documents
3. Create an API key under Settings

R2R (SciPhi)

Best for: Production-grade RAG with knowledge graphs

• MIT license
• REST-first API design
• Knowledge graph support
• Multimodal ingestion (text, images, audio)

Quick Start:

git clone https://github.com/SciPhi-AI/R2R.git
cd R2R
docker compose -f compose.full.yaml --profile postgres up -d

Default URL: http://localhost:7272

Custom RAG API

You can connect any RAG system that provides a REST API with:

• POST /search endpoint accepting {"query": "...", "top_k": 5}
• GET /collections endpoint listing available document collections
• GET /health endpoint for connection checking

Setup Guide

Method 1: Using Presets (Recommended)

1. Go to Admin > AI Settings > MCP Marketplace
2. Find the Knowledge Base (RAG) Presets section at the top
3. Click on your preferred provider card (e.g., RAGFlow)
4. Fill in the URL and authentication details
5. Click Save
6. The system will auto-discover the search_knowledge and list_knowledge_collections tools

Method 2: Manual Setup

1. Go to Admin > AI Settings > MCP Marketplace
2. Click Add Integration
3. Select RAG Knowledge Base as the source type
4. Enter your RAG server URL
5. Configure authentication:
   • RAGFlow: Bearer token (API key from RAGFlow settings)
   • AnythingLLM: Bearer token (API key from developer settings)
   • Dify: Bearer token (API key from Dify settings)
   • R2R: Bearer token or API key
   • Custom: Depends on your API
6. Set the Description to include the provider name (e.g., “ragflow”, “anythingllm”) for automatic provider detection
7. Click Save and then Check Connection

Binding to AI Features

After adding a knowledge base:

1. Open any AI feature (e.g., Mock Builder, API Tester)
2. Click the gear icon to open AI Settings
3. In the MCP Tools section, find your knowledge base
4. Check the search_knowledge tool
5. The AI agent will now automatically search your knowledge base when relevant

For a detailed guide on AI settings, MCP tool selection, and custom prompts, see the AI Features documentation.

Use Cases

Mock Generation with Real API Specs

Upload your OpenAPI/Swagger specs to the knowledge base. When you ask the AI to create a mock:

User: "Create a mock for the payment service"
Agent: [searches knowledge base for "payment service API"]
Agent: [finds OpenAPI spec with endpoints, schemas, examples]
Agent: [creates accurate mock with real data structures]

Report Generation in Company Format

Upload report templates and past reports:

User: "Generate a performance test report"
Agent: [searches for "performance test report template"]
Agent: [finds company template with required sections]
Agent: [generates report in the correct format]

Onboarding and Documentation

Upload internal documentation, wikis, and guides:

User: "How do I configure stores in Mockarty?"
Agent: [searches knowledge base for "store configuration"]
Agent: [finds internal guide with team-specific conventions]
Agent: [responds with accurate, team-specific instructions]

Coding Standards Enforcement

Upload coding standards and style guides:

User: "Generate a test script for the orders API"
Agent: [searches for "test script standards"]
Agent: [applies naming conventions, required assertions, etc.]

Architecture

The RAG adapter in the MCP Marketplace:

1. Auto-detects the provider type from the server name/description
2. Uses provider-specific API endpoints for search and listing
3. Formats responses with source attribution (document name, similarity score)
4. Supports per-user authentication via header overrides

Tips

• Start with RAGFlow or AnythingLLM for the simplest setup experience
• Use specific collection names in searches for better precision
• Upload focused documents rather than entire knowledge dumps
• Keep documents up to date — RAG results are only as good as the source data
• Use header overrides if different users need different API keys for the RAG system
• Monitor the Token Budget in AI settings – knowledge base results consume context tokens

Troubleshooting

Connection check fails

• Verify the RAG server is running: curl http://your-rag-server:port/health
• Check that the URL in Mockarty matches the RAG server’s actual address (watch for Docker networking issues – see the note about localhost vs Docker service names)
• Ensure the API key is correct and has not expired

AI agent does not use the knowledge base

• Make sure the search_knowledge tool is enabled in the AI settings (gear icon) for the specific feature you are using
• Check that the knowledge base contains documents relevant to your query – try searching directly in the RAG provider’s UI first
• The AI decides when to search based on relevance. If your question is generic, the AI may skip the search. Be specific in your prompts.

Poor search results

• Upload smaller, focused documents rather than large monolithic files
• Use meaningful file names – some providers index them
• If using RAGFlow, experiment with different chunking strategies in the dataset settings
• Ensure documents are in a supported format (PDF, DOCX, TXT, Markdown are universally supported)