SQLite MCP Server

The SQLite MCP server is maintained by Anthropic. It exposes SQL execution and schema inspection over MCP. AgentSUMO uses it to let the Planner Agent answer analytical questions about simulation results through natural-language → SQL translation against simulations.db.

Official documentation

modelcontextprotocol/servers-archived — SQLite

github.com/modelcontextprotocol/servers-archived/tree/main/src/sqlite

https://github.com/modelcontextprotocol/servers-archived/tree/main/src/sqlite

Tools used by AgentSUMO

The server exposes six tools across three categories, matching the layout in Figure 2 of the paper.

Category	Tool	Purpose
Query	read_query	Execute a SELECT query and return rows.
Query	write_query	Execute INSERT, UPDATE, or DELETE queries.
Query	create_table	Create new tables in the database.
Schema	list_tables	List the tables in simulations.db.
Schema	describe_table	Inspect a table’s schema.
Analysis	append_insight	Persist intermediate analytical findings as session memos.

How AgentSUMO uses it

Four concrete patterns:

1. Natural-language analytics. Questions like “Which road had the worst congestion?” or “Compare CO₂ emissions across the three EV scenarios” are translated to SQL by the LLM and executed through read_query.

2. Schema introspection during reasoning. list_tables and describe_table let the agent inspect the database before constructing a query — useful when it has been a few turns since the last xml_to_sqlite_tool call.

3. Materializing derived tables. write_query and create_table let the agent persist a per-corridor aggregate, a pre-joined view, or any other derived structure that subsequent questions reuse without rerunning the underlying joins.

4. Session-level memos. append_insight lets the agent record intermediate findings that survive across tool calls within the same conversation.

The same database holds all simulation runs in a single file, keyed by a unique simulation_id. This is what enables cross-scenario relational querying through ordinary SQL JOIN — see Database for the relational model.