Learn about how the Exasol MCP server works together with Exasol agent skills.
The Exasol MCP Server gives AI agents direct access to your Exasol database through the Model Context Protocol (MCP). Agents can discover schemas, inspect table structures, search for database objects, and run SQL queries, all without custom integration code.
This article explains how the MCP server works as an agent tool and how it pairs with Exasol agent skills to create agents that understand both Exasol’s SQL dialect and your live database.
When an agent connects to the Exasol MCP Server, it gains access to a set of capabilities that let it explore and query your database:
For the full list of tools, see the MCP Server tool list on GitHub.
These capabilities follow the MCP standard, so any MCP-compatible client can use them. This includes Claude Desktop, Claude Code, Cursor, Windsurf, and other tools that support MCP server connections.
The MCP server and agent skills solve two different problems:
An agent with only the MCP server can interact with your database, but it may generate SQL that does not account for Exasol-specific syntax (for example, DISTRIBUTE BY and PARTITION BY clauses, or Exasol’s reserved keywords). An agent with only skills has Exasol knowledge but cannot see or query your actual data.
When you combine MCP server and agent skills, the agent can inspect your schema through the MCP server and then apply its Exasol SQL knowledge from the skills to write correct, optimized queries.
The following published skills cover the core areas of Exasol development:
| Skill | What it teaches the agent |
|---|---|
exasol-database
|
Work with Exasol databases: queries, data loading, SQL quirks |
exasol-udfs
|
Build UDFs in Python, Java, Lua, or R, and package them into SLCs |
exasol-bucketfs
|
Manage files in BucketFS |
To give your agent Exasol knowledge alongside database access, install the agent skills:
curl -fsSL https://raw.githubusercontent.com/exasol-labs/exasol-agent-skills/main/install.sh | shFor full setup instructions (prerequisites, configuration, deployment modes), see Connect AI assistants (MCP Server).
This example shows how an agent with both the MCP server and agent skills handles the question: "What are the top 10 customers by total order value?".
Step 1: Schema discovery.
The agent uses the MCP server to list the schemas in your database and identify the one that contains sales data (for example, SALES).
Step 2: Table inspection.
The agent describes the tables in the SALES schema through the MCP server. It finds a CUSTOMERS table and an ORDERS table, and reads their column definitions.
Step 3: Query construction.
The agent writes a SQL query joining CUSTOMERS and ORDERS. Because it has the exasol-sql skill loaded, it uses Exasol-compatible syntax: correct join syntax, proper handling of reserved keywords, and Exasol-specific function names where they differ from standard SQL.
Step 4: Query execution.
The agent sends the query to the MCP server for execution and receives the result set.
Step 5: Response.
The agent presents the results in a readable format, explaining the data and noting any caveats about the query.
The entire interaction happens within the agent’s conversation. You ask a question in plain English and get back both the answer and the SQL the agent used, which you can verify and reuse.