AI Agent toolkit for VideoDB
llms.txt >> llms-full.txt
MCP

The VideoDB Agent Toolkit exposes VideoDB context to LLMs and agents. It enables integration to AI-driven IDEs like Cursor, chat agents like Claude Code etc. This toolkit automates context generation, maintenance, and discoverability. It auto-syncs SDK versions, docs, and examples and is distributed through MCP and llms.txt

The toolkit offers context files designed for use with LLMs, structured around key components:

llms-full.txt — Comprehensive context for deep integration.

llms.txt — Lightweight metadata for quick discovery.

MCP (Model Context Protocol) — A standardized protocol.

These components leverage automated workflows to ensure your AI applications always operate with accurate, up-to-date context.

llms-full.txt consolidates everything your LLM agent needs, including:

• Comprehensive VideoDB overview.

• Complete SDK usage instructions and documentation.

• Detailed integration examples and best practices.

Real-world Examples:

• VideoDB's Director code-assistant agent (View Implementation )
• VideoDB's Discord Bot to power customer support and community help (View Implementation )
• Integrate llms-full.txt directly into your LLM-powered workflows, agent systems, or AI coding environments.

A streamlined file following the Answer.AI llms.txt proposal. Ideal for quick metadata exposure and LLM discovery.

ℹ️ Recommendation: Use llms.txt for lightweight discovery and metadata integration. Use llms-full.txt for complete functionality.

The VideoDB MCP Server connects with the Director backend framework, providing a single tool for many workflows. For development, it can be installed and used via uvx for isolated environments. For more details on MCPs, please visit here

Install uv

We need to install uv first.

For macOS/Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh 

For Windows:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" 

You can also visit the installation steps of uv for more details here

Run the MCP Server

You can run the MCP server using uvx using the following command

uvx videodb-director-mcp --api-key=VIDEODB_API_KEY 

Update VideoDB Director MCP package

To ensure you're using the latest version of the MCP server with uvx, start by clearing the cache:

uv cache clean 

This command removes any outdated cached packages of videodb-director-mcp, allowing uvx to fetch the most recent version.

If you always want to use the latest version of the MCP server, update your command as follows:

uvx videodb-director-mcp@latest --api-key=<VIDEODB_API_KEY> 

LLM context files in VideoDB are modular, automatically generated, and continuously updated from multiple sources:

• Instructions — Best practices and prompt guidelines View »

• SDK Context — SDK structure, classes, and interface definitions View »

• Docs Context — Summarized product documentation View »

• Examples Context — Real-world notebook examples View »

• Managed through GitHub Actions for automated updates.
• Triggered by changes to SDK repositories, documentation, or examples.
• Maintained centrally via a config.yaml file.

Automatic context generation ensures your applications always have the latest information:

• Automatically generates documentation from SDK repo updates.
• Uses Sphinx for Python SDKs.

• Scrapes and summarizes documentation using FireCrawl and LLM-powered summarization.

• Converts and summarizes notebooks into practical context examples.

• Combines all sub-components into unified llms-full.txt.
• Generates standards-compliant llms.txt.
• Updates documentation with token statistics for transparency.

The config.yaml file centralizes all configurations, allowing easy customization:

• Inclusion & Exclusion Patterns for documentation and notebook processing
• Custom LLM Prompts for precise summarization tailored to each document type
• Layout Configuration for combining context components seamlessly

config.yaml > llms_full_txt_file defines how llms-full.txt is assembled:

llms_full_txt_file: input_files: - name: Instructions file_path: "context/instructions/prompt.md" - name: SDK Context file_path: "context/sdk/context/index.md" - name: Docs Context file_path: "context/docs/docs_context.md" - name: Examples Context file_path: "context/examples/examples_context.md" output_files: - name: llms_full_txt file_path: "context/llms-full.txt" - name: llms_full_md file_path: "context/llms-full.md" layout: |  {{FILE1}}   {{FILE2}}   {{FILE3}}   {{FILE4}} 

• Automate Context Updates: Leverage GitHub Actions to maintain accuracy.
• Tailored Summaries: Use custom LLM prompts to ensure context relevance.
• Seamless Integration: Continuously integrate with existing LLM agents or IDEs.

By following these practices, you ensure your AI applications have reliable, relevant, and up-to-date context—critical for effective agent performance and developer productivity.

Clone the toolkit repository and follow the setup instructions in config.yaml to start integrating VideoDB contexts into your LLM-powered applications today.

Explore further:

• VideoDB SDK
• Documentation
• Cookbook Examples