Drill-MCP Server

The official MCP server for Apache Drill. It lets an MCP client run read-only (and narrowly, explicitly allow-listed write) SQL against a Drill cluster, and inspect schemas, storage plugins, and cluster/query state. It does not implement Drill administration (ALTER SYSTEM, storage-plugin management) — see “Safety model” below.

Install

pip install drill-mcp             # REST backend
pip install drill-mcp[jdbc]       # adds the JDBC backend (needs a JVM)
pip install drill-mcp[kerberos]   # adds SPNEGO auth for the REST backend

The base install has no JVM dependency: the default REST backend talks to Drill over plain HTTP via httpx. The jdbc extra pulls in jaydebeapi/JPype1 and is only needed if you select backend: jdbc (see “Backends” below).

Quickstart

drill-mcp --url http://localhost:8047
# or
drill-mcp --config /etc/drill-mcp/drill.yaml

drill-mcp --help lists every CLI flag.

{
  "mcpServers": {
    "drill": {
      "command": "drill-mcp",
      "args": ["--config", "/etc/drill-mcp/drill.yaml"]
    }
  }
}

Configuration

Configuration is merged from three sources, each overriding the last: config file → DRILL_* environment variables → CLI flags. Because the environment wins over the file, a stale exported DRILL_PASSWORD silently overrides a password: set in the config file — unset it if the file is meant to be authoritative.

There is no --user/--password CLI flag and no tool accepts a credential as an argument; user/password can only be set via the config file or DRILL_USER/DRILL_PASSWORD.

Every config key, its default, its environment variable, and its CLI flag are documented in docs/configuration.md. drill.example.yaml is a complete example file with every key present.

Tools

Tool	Description
`run_query`	Run one SQL statement
`list_schemas`	List visible schemas
`list_tables`	List tables in a schema
`describe_table`	Column names, types, and nullability where the plugin reports it
`list_storage_plugins`	Plugin configs, secrets redacted
`cluster_status`	Drillbit membership and status
`list_profiles`	Recent and running queries
`get_profile`	Full profile for one query id
`cancel_query`	Cancel a running query

list_storage_plugins, cluster_status, list_profiles, get_profile, and cancel_query are management tools: they require the REST backend and raise a clear ToolError on the JDBC backend rather than failing silently or returning empty data.

Full parameters, return shapes, errors, and per-tool surprises are in docs/tools.md.

Safety model

Writes are denied by default. CREATE TABLE AS, CREATE VIEW, DROP TABLE, and DROP VIEW are permitted only into plugins listed in writable_plugins. INSERT, UPDATE, DELETE, MERGE, ALTER, USE, and REFRESH are always rejected, no matter what writable_plugins contains.
ALTER SYSTEM and storage-plugin create/update/delete are not implemented at all. There is no flag, config key, or code path that turns them on — there is nothing for such a tool to call.
Exactly one statement per call, checked with a real SQL parser (sqlglot's Drill dialect), not a regex, so a write hidden in a comment, a string literal, or a stacked statement does not slip through. Anything that fails to parse — including a tokenizer failure — is rejected.
EXPLAIN recurses. The guard strips a leading EXPLAIN or EXPLAIN PLAN FOR and re-checks the remaining statement against the same allowlist and hidden-schema rules, with bounded recursion depth.
hidden_schemas filters more than schema listings. Beyond refusing queries against a hidden schema, it filters the result rows of every SHOW command (not only SHOW SCHEMAS), because Drill's SHOW grammar gives sqlglot no reliable way to identify which spelling lists schemas. This is a deliberate fail-closed trade: a table or file whose name happens to match a hidden-schema prefix is incidentally hidden too. See docs/tools.md for the full explanation. Hiding INFORMATION_SCHEMA does not break the metadata tools, which query it internally regardless and simply omit hidden entries from their output.
Secrets are always redacted from list_storage_plugins output. This is not configurable off.
Results are capped by max_rows; a caller may lower the cap for a single call but never raise it above the configured limit. A truncated result is marked "truncated": true.

Backends

REST (default, backend: rest): talks to a Drillbit's HTTP endpoints over httpx. Works out of the box, no JVM required.
JDBC (backend: jdbc, needs the jdbc extra): uses jaydebeapi/JPype1 against Drill‘s JDBC driver. It exists mainly because Kerberos is materially less painful through the Drill JDBC driver than through Python SPNEGO. It needs a JVM and jdbc_driver_path pointing at drill-jdbc-all.jar. Management tools are unavailable on this backend, since Drill’s management API is REST-only, and run_query never returns a query_id on this backend.

Wire-level behavior — identifier quoting, INFORMATION_SCHEMA queries, and the per-plugin-type column-discovery strategy in describe_table — follows sqlalchemy-drill, the most complete and maintained reference for talking to Drill from Python. PyDrill was evaluated and not adopted: it has no form-based login, no Kerberos support, and its last release was in 2018.

Development

pip install -e ".[dev]"
pytest

The test suite needs no live Drill cluster and no JVM: Drill's wire protocol is mocked at the transport boundary, so pytest runs anywhere Python does.

Credits and license

Wire-level behavior follows sqlalchemy-drill.

Licensed under the Apache License, Version 2.0. See NOTICE.

Apache drill

Branches