
Apache Doris MCP Server
MCPApache Doris's official MCP: gives agents read-only access to explore table metadata, run and explain SQL, diagnose slow queries, and check cluster status, load jobs and data governance info. Needs Python 3.12+ and Doris 2.0 or later.
Apache Doris MCP Server
Apache Doris MCP Server exposes read-only Apache Doris capabilities to MCP
Hosts and AI agents over MCP 2026-07-28. Version 1.0 replaces a large flat
tool surface with eight stable domains and fifty-five progressively disclosed
child capabilities, while keeping runtime availability, authorization, input
schemas, output schemas, and failure behavior explicit.
Release status
The package version is 1.0.0. MCP 2026-07-28 protocol compatibility on
master is Generally Available (GA) on Streamable HTTP and stdio.
This GA statement is scoped to protocol compatibility; the Python package
classifier remains Beta, and the documented deployment limits still apply.
Before upgrading, read the 1.0 release notes, the 1.0 migration guide, and the generated 8-domain/55-child registry. The detailed release record is Issue #189.
Architecture at a glance
MCP Host
-> stdio or Streamable HTTP
-> transport security and authentication
-> MCP protocol validation and authorization
-> stable domain discovery
-> route-aware Doris capability detection
-> exact child dispatch and read-only runtime
-> request-specific Doris route and RBAC
-> bounded, schema-validated result
The default hierarchical mode exposes these domains:
| Domain | Children | Responsibility |
|---|---|---|
doris_catalog | 5 | catalogs, databases, tables, table context, size |
doris_query | 7 | query, explain, profile, diagnosis, slow queries, explicit ADBC |
doris_cluster | 11 | nodes, tasks, metrics, memory, cache, compaction, workloads |
doris_pipeline | 5 | ingestion, materialized views, freshness, dependencies |
doris_search | 4 | text/vector/hybrid search, analyzers, indexes, diagnosis |
doris_governance | 8 | quality, storage, lineage, audit, UDFs, auth mapping |
doris_lakehouse | 3 | external catalogs, lakehouse tables, Variant |
doris_semantic | 12 | optional Apache Ossie grounding and MetricFlow consumption |
Call a domain with {} to discover its authorized children and exact schemas.
Call the same domain again with child_tool, arguments, and the returned
manifest_version. Hosts that cannot use progressive disclosure may set
MCP_TOOL_EXPOSURE_MODE=flat before startup; this exposes the same 55 children
under collision-free formal names and does not restore pre-1.0 aliases.
See Architecture, Request lifecycle, and Tool domains.
Quick start
Requirements:
- Python 3.12 or later;
- Apache Doris 2.0.0 or later;
- network access to the Doris FE MySQL endpoint, normally port
9030.
Install the pinned release:
pip install doris-mcp-server==1.0.0
doris-mcp-server starts the Server. doris-mcp-client is a separate client;
the two commands are not interchangeable.
Configure a Doris route:
export DORIS_HOST=127.0.0.1
export DORIS_PORT=9030
export DORIS_USER=mcp_reader
export DORIS_PASSWORD='replace-me'
export DORIS_DATABASE=information_schema
Start Streamable HTTP on loopback:
doris-mcp-server \
--transport http \
--host 127.0.0.1 \
--port 3000
Endpoints:
- MCP:
POST http://127.0.0.1:3000/mcp - legacy MCP (opt-in):
POST http://127.0.0.1:3000/mcp/legacy - liveness:
GET http://127.0.0.1:3000/live - Doris-backed readiness:
GET http://127.0.0.1:3000/ready
Hosts limited to handshake-era Streamable HTTP, including Dify 1.16.1 with
MCP 2025-06-18, must set ENABLE_LEGACY_HTTP_ADAPTER=true and connect to
/mcp/legacy. The adapter changes only the protocol boundary; it preserves the
same 1.0 tools, authorization, capability gates, and read-only execution.
Or run stdio for a local Host:
doris-mcp-server --transport stdio
See the complete Quick start and Host integration guide.
Security boundary
- The built-in 1.0 catalog is read-only;
doris_adminis reserved and not registered. - Static tokens, JWT, external OAuth/OIDC, and Doris-backed OAuth are supported under mutually validated configuration boundaries.
- Domain discovery and child execution use exact authorization identifiers.
- Doris RBAC remains the final authority for visible objects and data.
- SQL shape, identifiers, parameters, timeout, rows, bytes, and result schemas are bounded before data leaves the Server.
- Secrets and backend errors are redacted from public results and logs.
- Non-loopback HTTP requires authentication unless an explicit dangerous development override is enabled.
Read the Security and permission model and the Doris fine-grained access guide.
Reliability boundary
The Server uses deterministic manifests and errors, signed expiring cursors,
route-aware capability snapshots, bounded stale fallback, request-specific
connection routing, multi-FE failover, liveness/readiness separation, output
Schema validation, and sanitized trace propagation. Unsupported or
misconfigured capabilities remain discoverable with callable=false and fail
closed when called.
Current limits include process-local Doris-backed OAuth, explicit-only ADBC that is disabled by default and fail-closed on token-bound routes, optional read-only Ossie grounding, an optional MetricFlow compiler sidecar whose SQL must execute through the bounded MCP query runtime, and best-effort native lineage delivery. See Reliability and limits.
Documentation
The root README is intentionally an entry point. The bilingual documentation system is indexed at:
Primary guides:
- Architecture
- Request and data flow
- Tool domains
- Capability availability
- Doris version capability matrix
- MetricFlow integration
- MCP 2026-07-28 contract
- Security model
- Deployment
- Reliability and limits
- Troubleshooting
- Configuration reference
- Host integrations
- Custom tool providers
- Contributing
Development
git clone https://github.com/apache/doris-mcp-server.git
cd doris-mcp-server
uv sync --group dev
uv run pytest
Generated artifacts must remain synchronized:
uv run python generate_tool_catalog.py --check
uv lock --check
See Contributing and verification.
License
Apache License 2.0. See LICENSE.txt and NOTICE.
Recursos relacionados

Datasette
CLIPublishes SQLite databases as browsable, queryable web pages and JSON APIs, with SQL queries, metadata and plugins, and deploys to Cloud Run and other platforms with datasette publish. Great for data exploration and data journalism. Needs Python 3.10+.
simonw/datasette11.5k
sqlite-utils
CLIA CLI and Python library by Simon Willison: imports JSON and CSV straight into SQLite with automatic table creation, runs in-memory SQL queries over files, and supports full-text search and table transformations.
simonw/sqlite-utils2.2k
MySQL MCP Server
MCPGives agents controlled access to MySQL: list databases and tables, inspect schemas and sample data, and run SQL, with multi-database mode, SSL and SSH tunnels. Configure your own database user, ideally with least privilege.
designcomputer/mysql_mcp_server1.4k
ClickHouse MCP Server
MCPRuns SQL queries against ClickHouse and browses databases and table schemas, read-only by default; can also enable embedded chDB to query data directly in files and URLs. Maintained by ClickHouse.
ClickHouse/mcp-clickhouse888
DuckDB
CLIIn-process analytical SQL database. Its CLI queries CSV, Parquet and JSON files directly with SQL, no import or server needed, so agents can sample and aggregate large files. Officially maintained.
duckdb/duckdb42.2k
SQL Queries
SkillAnthropic's official data skill that guides agents to write, translate and optimize SQL across Snowflake, BigQuery, Databricks, PostgreSQL and other warehouse dialects, including CTEs, window functions and analytical aggregations.
anthropics/knowledge-work-plugins29.9k