DataAgentConnector

MagnusS04stars

Blueprint for spinning up read-only SQL data agents from any SQLAlchemy-compatible database. Indexes schema and column content into LanceDB and exposes MCP + REST tools so LLMs and UIs can safely expl

Databases

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# DataAgentConnector

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 4
Tools: —
Uptime: —

Overview

About DataAgentConnector

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install DataAgentConnector?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does DataAgentConnector need an API key?

No required environment variable is listed on this record.

What transport does DataAgentConnector use?

DataAgentConnector runs locally over stdio.

When was DataAgentConnector last checked?

Last checked 2026-10-28. Blueprint for spinning up read-only SQL data agents from any SQLAlchemy-compatible database. Indexes schema and column content into LanceDB and exposes MCP + REST tools so LLMs and UIs can safely explore your data.

From the publisher

Documentation

Data Agent Connector

This provides a blueprint to go from database connection string to up and running SQL Data Agent in seconds. Connect to any database that can be used with SQLAlchemy (might need to install specific engine). Builds search and convenience tools on top of a DB, made for SQL agents to interact with through the MCP protocol. Plus mirrored REST endpoint that can be connected to UIs etc.

Key Features

Read-only SQL gateway: SQLAlchemy engines are locked to safe commands defined in [tool.dac.settings.allowed_sql_commands].
Automatic metadata: LLM agents summarise tables; LanceDB stores the summaries.
Column-content retrieval: Distinct textual values are sampled, filtered, and indexed with LanceDB BM25 for direct content search in columns.
MCP + REST: /mcp serves FastMCP, while /widgets/* exposes REST endpoints for UI integration with OpenBB (customize this to your preferred UI).
Config-driven: databases.toml declares available data sources, pyproject.toml under [tool.dac.settings] for runtime settings and .env (or environment variables) configures the LLM provider.

MCP (Mounted at /mcp)

Tool	Summary
get_databases	Lists registered databases and descriptions.
show_tables / show_views	Enumerates tables/views with cached annotations where available.
describe_tables / describe_view	Returns DDL-like metadata or view SQL.
get_distinct_values	Pulls sample categorical values (limit enforced).
preview_table	Returns first rows of non-binary columns.
find_relevant_columns_and_content	Typo-tolerant BM25 search over distinct textual values; returns the matching values per column.
query_database	Executes read-only SQL with a configurable row cap (mcp_query_limit).
join_path	Suggests shortest join sequences or Steiner-tree paths across tables.

Getting Started

Clone the repository:

git clone https://github.com/MagnusS0/DataAgentConnector.git
cd DataAgentConnector

Install dependencies:
```
uv sync --group ai
```

Configure your databases in databases.toml:

[databases.my_database]
url = "sqlite:///path/to/your/database.db"
description = "My local SQLite database"

[databases.another_database]
url = "postgresql://user:password@localhost:5432/another_database"
description = "Another PostgreSQL database"

Set up your LLM provider in .env:

LLM_API_KEY=your_api_key_here
LLM_MODEL_NAME=default-model
LLM_BASE_URL=https://api.your-llm-provider.com

Run the application:
```
uv run uvicorn app.main:app --reload
```

Project Structure

DataAgentConnector/
├── app/
│   ├── agents/
│   ├── core/
│   ├── domain/
│   ├── interfaces/
│   ├── models/
│   ├── schemas/
│   ├── repositories/
│   ├── services/
│   └── main.py
├── databases.toml
├── pyproject.toml
├── .env
└── README.md

Indexing & Metadata Pipeline

Column extraction (app/domain/extract_colum_content.py) samples distinct textual values while filtering binary, numeric, or overly long fields; tunable via tool.dac.settings.fts_extraction_options.
FTS indexing (app/services/indexing_service.py) persists values into LanceDB tables named column_contents_<database> and builds BM25 indexes that score each distinct value on its own. Searches boost exact matches over typo-tolerant (one edit) matches and return the matching values per column.
Annotation workflow (app/services/annotation_service.py) runs LLM prompts with table metadata, previews, and sampled values (schema hashes used to skip already processed tables).

FK Graph & Join Paths

Foreign key constraints are analyzed to build a cached CSR adjacency matrix (app/domain/fk_analyzer.py) where tables are nodes and FKs are edges. For two tables, BFS finds the shortest join sequence. For 3+ tables, an approximate Steiner tree (MST on all-pairs distances) computes the minimal spanning network, returning ordered JoinStep objects with FK column mappings.

This allows agents to request optimal join paths across multiple tables when formulating SQL queries. Even when there is no direct foreign key relationship defined in the database schema.

Stats for the interested user

Indexing and annotating all of BIRD-SQL training databases (69 databases) results in:

Table annotations stored successfully in ~200 seconds
Content FTS indices created successfully in ~5 seconds

Hardware: Intel i9-14900K, 64GB RAM, RTX 3090 running Menlo/Jan-nano (4B params) using vLLM

DataAgentConnector

MagnusS04stars

Databases

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# DataAgentConnector

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 4
Tools: —
Uptime: —

Overview

About DataAgentConnector

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install DataAgentConnector?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does DataAgentConnector need an API key?

No required environment variable is listed on this record.

What transport does DataAgentConnector use?

DataAgentConnector runs locally over stdio.

When was DataAgentConnector last checked?

From the publisher

Documentation

Data Agent Connector

Key Features

Read-only SQL gateway: SQLAlchemy engines are locked to safe commands defined in [tool.dac.settings.allowed_sql_commands].
Automatic metadata: LLM agents summarise tables; LanceDB stores the summaries.
Column-content retrieval: Distinct textual values are sampled, filtered, and indexed with LanceDB BM25 for direct content search in columns.
MCP + REST: /mcp serves FastMCP, while /widgets/* exposes REST endpoints for UI integration with OpenBB (customize this to your preferred UI).
Config-driven: databases.toml declares available data sources, pyproject.toml under [tool.dac.settings] for runtime settings and .env (or environment variables) configures the LLM provider.

MCP (Mounted at /mcp)

Tool	Summary
get_databases	Lists registered databases and descriptions.
show_tables / show_views	Enumerates tables/views with cached annotations where available.
describe_tables / describe_view	Returns DDL-like metadata or view SQL.
get_distinct_values	Pulls sample categorical values (limit enforced).
preview_table	Returns first rows of non-binary columns.
find_relevant_columns_and_content	Typo-tolerant BM25 search over distinct textual values; returns the matching values per column.
query_database	Executes read-only SQL with a configurable row cap (mcp_query_limit).
join_path	Suggests shortest join sequences or Steiner-tree paths across tables.

Getting Started

Clone the repository:

git clone https://github.com/MagnusS0/DataAgentConnector.git
cd DataAgentConnector

Install dependencies:
```
uv sync --group ai
```

Configure your databases in databases.toml:

[databases.my_database]
url = "sqlite:///path/to/your/database.db"
description = "My local SQLite database"

[databases.another_database]
url = "postgresql://user:password@localhost:5432/another_database"
description = "Another PostgreSQL database"

Set up your LLM provider in .env:

LLM_API_KEY=your_api_key_here
LLM_MODEL_NAME=default-model
LLM_BASE_URL=https://api.your-llm-provider.com

Run the application:
```
uv run uvicorn app.main:app --reload
```

Project Structure

DataAgentConnector/
├── app/
│   ├── agents/
│   ├── core/
│   ├── domain/
│   ├── interfaces/
│   ├── models/
│   ├── schemas/
│   ├── repositories/
│   ├── services/
│   └── main.py
├── databases.toml
├── pyproject.toml
├── .env
└── README.md

Indexing & Metadata Pipeline

Column extraction (app/domain/extract_colum_content.py) samples distinct textual values while filtering binary, numeric, or overly long fields; tunable via tool.dac.settings.fts_extraction_options.
FTS indexing (app/services/indexing_service.py) persists values into LanceDB tables named column_contents_<database> and builds BM25 indexes that score each distinct value on its own. Searches boost exact matches over typo-tolerant (one edit) matches and return the matching values per column.
Annotation workflow (app/services/annotation_service.py) runs LLM prompts with table metadata, previews, and sampled values (schema hashes used to skip already processed tables).

FK Graph & Join Paths

This allows agents to request optimal join paths across multiple tables when formulating SQL queries. Even when there is no direct foreign key relationship defined in the database schema.

Stats for the interested user

Indexing and annotating all of BIRD-SQL training databases (69 databases) results in:

Table annotations stored successfully in ~200 seconds
Content FTS indices created successfully in ~5 seconds

Hardware: Intel i9-14900K, 64GB RAM, RTX 3090 running Menlo/Jan-nano (4B params) using vLLM