AnyDB

You have an MCP server called AnyDB MCP Server that lets you interact with SQLite databases through natural language. It combines AI-powered SQL generation with safe read/write tooling, supports dynamic schema evolution, and integrates with MCP clients for seamless conversational data operations and document-aware retrieval.

How to use

You connect an MCP client such as Claude Desktop to the AnyDB MCP Server to start working with your SQLite data using everyday language. You can ask it to query tables, insert or update records, and perform schema changes automatically as your data evolves. You also gain access to a vector database for semantic search and a web-scraping workflow to build a knowledge base you can query with natural language.

How to install

Follow these concrete steps to set up and run the server locally.

# 1. Clone the project
git clone https://github.com/iamayuppie/AnyDbApp.git
cd AnyDbApp

# 2. Install Python dependencies
pip install -r requirements.txt

# 3. Start Ollama and pull a model
ollama serve --port 1434
ollama pull llama3.1  # or your preferred model

# 4. Run the MCP server
python main.py

Configuration and usage notes

The server includes an MCP integration workflow that lets Claude Desktop connect via a local stdio endpoint. When you set up Claude Desktop, add a MCP server entry that points to the stdio runner, and provide the necessary environment path so the server can be located.

Troubleshooting

If the server won’t start, verify that Ollama is running on port 1434. If tools don’t appear in Claude Desktop, ensure the MCP config path is correct and restart Claude Desktop. For SQL errors, check that table names exist and that your natural language descriptions map to valid SQL. If the Ollama connection fails, confirm the model is installed and accessible.

Notes on integration and runtime

This server is designed to run as a local stdio MCP server. The recommended run flow uses Python to start the server, while Ollama provides the AI model for SQL generation. The default SQLite database is anydb.sqlite and is created automatically in the server directory. The vector store uses ChromaDB to persist embeddings for semantic search and RAG.

Tooling and capabilities overview

You can exercise a variety of database and vector operations through natural language prompts. The server includes tools for querying, inserting, updating, and deleting data, as well as creating new tables and executing raw SQL. It also supports vector-based document ingestion, semantic search over files, as well as web scraping to enrich your knowledge base.

Available tools

query_entity

Query any table with natural language instructions.

insert_entity

Insert records into any table using natural language descriptions and auto-schema evolution.

update_entity

Update records with conditions; new fields can be added automatically.

delete_entity

Delete records with optional conditions.

create_table

Create new tables with AI-generated schemas.

sql_query

Execute raw SQL SELECT queries directly.

sql_execute

Execute raw SQL modification queries (INSERT/UPDATE/DELETE/CREATE).

add_file_to_vector_db

Add a file to the vector database for semantic search and RAG.

search_vector_db

Semantic search in the vector database.

list_vector_files

List all files stored in the vector database.

remove_file_from_vector_db

Remove a file from the vector database.

scrape_url

Scrape a web page and store content for semantic search.

query_web_content

Query scraped web content with semantic search.

list_scraped_pages

List all scraped web pages with metadata.

remove_scraped_page

Remove a scraped page from the vector database.