Chat App MCP

🚀 MCP Server for LLM Agents

The MCP Server is the core of our solution. It enables the LLM Chat and Gitlab Agent services to interact with the infrastructure where the MCP Server is deployed.

On the MCP Server side, interactions with the following objects are implemented:

• Databases, all of which have drivers for sqlalchemy
• File storages (S3, SMB)
• Kafka
• Web interfaces

The following data formats are supported for file storages:

• CSV, TSV
• JSON, XML
• Parquet

Main actions on databases:

• Check database availability
• Check the existence of a schema.table
• Get DDL (table structure)
• Get a data sample to verify the table structure

Main actions on storages:

• Check storage availability
• Check directory availability and the existence of a file
• Read a file or part of it (for text formats)
• Get the schema for Parquet files

🚀 Quick Start

📦 Installation

Our project is prepared for containerization and includes a Dockerfile for building the image.

To run the service, the following environment variables need to be configured:

• HTTP_PROXY - Proxy server address if your environment doesn't have direct internet access
• HTTPS_PROXY - Proxy server address for HTTPS requests if your environment doesn't have direct internet access
• GITLAB_URL - Gitlab server address for API access
• GITLAB_TOKEN - A valid token with read and write access via API
• PROJECT_PATH - Path to the project where branches will be created
• AIRFLOW_URL - Apache Airflow server address
• AIRFLOW_USER - User with access to the Airflow server's REST API
• AIRFLOW_PASSWORD - User password

To build the image, specify all the necessary environment variables and start the build process:

$ git clone https://github.com/AnatoliyAksenov/chat-app-backend.git
$ cd chat-app-backend
$ docker build -t chat-app-backend:0.0.1 --build-arg HTTPS_PROXY=http://10.0.0.7:3128 \
--build-arg LDAP_SERVER=10.0.0.100 \
--build-arg LDAP_SERVER=1389 \
--build-arg POSTGRESQL_URL=postgres://10.0.0.7:5432/chat .

Also, for our test environment, we've configured the application's build and deployment via CI/CD.

📚 Documentation

🔧 Technical Details

Duplicated from the Gitlab Agent repository

The agent consists of two main modules:

• LLM Chat (user interface)
• LLM Gitlab Agent (service)

Also, the MCP Server with a set of tools for interacting with databases and storages is required for the services to work.

The creation of a new pipeline for data loading starts in the chat. The user provides all the necessary data for creating a new data load during the conversation with the agent. After collecting the necessary data, the chat agent creates an issue in the Gitlab application, which contains all the information for creating the pipeline. After the issue is created, Gitlab calls a webhook to the LLM Gitlab Agent and starts processing the issue. The Gitlab Agent processes the issue, creates a new branch in the designated project in the Gitlab repository, transfers all the generated code and documentation to it, creates a Merge Request, and completes its work.

When the Merge Request is accepted, the CI/CD for delivering changes to Airflow is triggered.

Related Projects

• MCP Server: https://github.com/AnatoliyAksenov/chat-app-mcp
• LLM Chat backend: https://github.com/AnatoliyAksenov/chat-app-backend
• LLM Chat frontend: https://github.com/AnatoliyAksenov/chat-app-frontend
• Gitlab Agent: https://github.com/AnatoliyAksenov/gitlab-agent

Also, we've prepared a deployed test environment.