greptimedb-mcp-server
A Model Context Protocol (MCP) server for GreptimeDB — an open-source observability database that handles metrics, logs, and traces in one engine.
Enables AI assistants to query and analyze GreptimeDB using SQL, TQL (PromQL-compatible), and RANGE queries, with built-in security features like read-only enforcement and data masking.
Features
Tools
| Tool | Description |
|---|
execute_sql | Execute SQL queries with format (csv/json/markdown) and limit options |
execute_tql | Execute TQL (PromQL-compatible) queries for time-series analysis |
query_range | Execute time-window aggregation queries with RANGE/ALIGN syntax |
search_table_semantics | Find tables by observability concept, ranked by matched terms; searches table names, semantic options, and entity declarations. schema selects the database to search |
query_semantic_graph | Query the semantic graph: summary (what it contains), entities (nodes), relationships (edges) over a required time window |
describe_table | Inspect a table profile: schema, semantic metadata, latest sample rows, and query guidance |
explain_query | Analyze SQL or TQL query execution plans (analyze=true for runtime stats; add verbose=true alongside analyze=true for per-partition scan metrics and index-pruning counters) |
health_check | Check database connection status and server version |
search_table_semantics and the semantic metadata in describe_table read information_schema.table_semantics. A table appears there when it carries a greptime.semantic.* option or a built-in convention derives an entity declaration for it; other tables are absent. The server reads the view's column list once per process and selects only the columns it exposes. entity_declarations requires GreptimeDB 1.3; on earlier versions it is reported as a missing column rather than as an empty declaration set.
Tool results are bounded by GREPTIMEDB_MAX_RESULT_BYTES (default 1048576, or 1 MiB). A query result over the budget returns fewer rows and reports it in truncated and truncation_reason; a result with no rows to shed is cut with a notice at the end. The budget is a backstop against a result too large to be consumed, so lower it if your MCP client rejects results before it is reached.
Cross-database queries
The server connects to one database, but reads are not restricted to it:
| Tool | How to reach another database |
|---|
execute_sql, query_range, describe_table | Qualify the table as schema.table |
execute_tql | metric{__schema__="other_db"}, which accepts = only |
search_table_semantics | Pass schema |
query_semantic_graph | Catalog-wide already; no scoping needed |
The connected account needs read permission on the target database. Use SHOW DATABASES to list them.
query_semantic_graph reads greptime_private.semantic_entities and greptime_private.semantic_relationships, which require GreptimeDB 1.3. At startup the server checks that both views exist, carry the columns it reads, and are readable by the connected account; when they are not, the tool is not offered and the reason is logged. Its time window is required and half-open, [start_time, end_time) over observed_at, and rows are aggregated across the 60-second observation buckets in that window.
Pipeline Management
| Tool | Description |
|---|
list_pipelines | List all pipelines or get details of a specific pipeline |
create_pipeline | Create a new pipeline with YAML configuration |
dryrun_pipeline | Test a pipeline with sample data without writing to database |
delete_pipeline | Delete a specific version of a pipeline |
Dashboard Management
| Tool | Description |
|---|
list_dashboards | List all Perses dashboard definitions |
create_dashboard | Create or update a Perses dashboard definition |
delete_dashboard | Delete a dashboard definition |
Resources & Prompts
- Resources: Browse tables via
greptime://<table>/data URIs
- Prompts: Built-in Jinja templates for common tasks —
pipeline_creator, log_pipeline, metrics_analysis, promql_analysis, trace_analysis, table_operation, schema_design_advisor, observability_correlation, ingestion_troubleshooting, query_performance_tuning
For LLM integration and prompt usage, see docs/llm-instructions.md.
These tools cover querying and managing data in an existing GreptimeDB. For deployment, server configuration, write protocols, pipeline syntax, schema design, and performance diagnosis, point the assistant at the GreptimeDB skills index at https://docs.greptime.com/SKILL.md.
Configuration
Environment Variables
GREPTIMEDB_HOST=localhost # Database host
GREPTIMEDB_PORT=4002 # MySQL protocol port (default: 4002)
GREPTIMEDB_USER=root # Database user
GREPTIMEDB_PASSWORD= # Database password
GREPTIMEDB_DATABASE=public # Database name
GREPTIMEDB_TIMEZONE=UTC # Session timezone
# Optional
GREPTIMEDB_HTTP_PORT=4000 # HTTP API port for pipeline/dashboard management
GREPTIMEDB_HTTP_PROTOCOL=http # HTTP protocol (http/https)
GREPTIMEDB_POOL_SIZE=5 # Connection pool size
GREPTIMEDB_MASK_ENABLED=true # Enable sensitive data masking
GREPTIMEDB_MASK_PATTERNS= # Additional patterns (comma-separated)
GREPTIMEDB_AUDIT_ENABLED=true # Enable audit logging
GREPTIMEDB_ALLOW_WRITE=false # Allow writes: DDL/DML plus pipeline and dashboard changes (DANGEROUS, local/test only)
GREPTIMEDB_MAX_RESULT_BYTES=1048576 # Byte budget for a single tool result (1 MiB)
# Transport (for HTTP server mode)
GREPTIMEDB_TRANSPORT=stdio # stdio, sse, or streamable-http
GREPTIMEDB_LISTEN_HOST=0.0.0.0 # HTTP server bind host
GREPTIMEDB_LISTEN_PORT=8080 # HTTP server bind port
GREPTIMEDB_ALLOWED_HOSTS= # DNS rebinding protection (comma-separated)
GREPTIMEDB_ALLOWED_ORIGINS= # CORS allowed origins (comma-separated)
CLI Arguments
greptimedb-mcp-server \
--host localhost \
--port 4002 \
--database public \
--user root \
--password "" \
--timezone UTC \
--pool-size 5 \
--mask-enabled true \
--allow-write false \
--transport stdio
HTTP Server Mode
For containerized or Kubernetes deployments:
# Streamable HTTP (recommended for production)
greptimedb-mcp-server --transport streamable-http --listen-port 8080
# SSE mode (legacy)
greptimedb-mcp-server --transport sse --listen-port 3000
DNS Rebinding Protection
By default, DNS rebinding protection is disabled for compatibility with proxies, gateways, and Kubernetes services. To enable it, use --allowed-hosts:
# Enable DNS rebinding protection with allowed hosts
greptimedb-mcp-server --transport streamable-http \
--allowed-hosts "localhost:*,127.0.0.1:*,my-service.namespace:*"
# With custom allowed origins for CORS
greptimedb-mcp-server --transport streamable-http \
--allowed-hosts "my-service.namespace:*" \
--allowed-origins "http://localhost:*,https://my-app.example.com"
# Or via environment variables
GREPTIMEDB_ALLOWED_HOSTS="localhost:*,my-service.namespace:*" \
GREPTIMEDB_ALLOWED_ORIGINS="http://localhost:*" \
greptimedb-mcp-server --transport streamable-http
If you encounter 421 Invalid Host Header errors, either disable protection (default) or add your host to the allowed list.
Security
Read-Only Database User (Recommended)
Create a read-only user in GreptimeDB using static user provider:
mcp_readonly:readonly=your_secure_password
Application-Level Security Gate
All queries go through a security gate that:
- Blocks: DROP, DELETE, TRUNCATE, UPDATE, INSERT, ALTER, CREATE, GRANT, REVOKE, EXEC, LOAD, COPY
- Blocks: Encoded bypass attempts (hex, UNHEX, CHAR)
- Allows: SELECT, SHOW, DESCRIBE, TQL, EXPLAIN, UNION
Write Mode (Disabled by Default)
The server is read-only by default. Write mode enables every operation that
changes stored state:
- destructive SQL (DDL/DML such as
CREATE, DROP, ALTER, INSERT, UPDATE,
DELETE) through execute_sql
create_pipeline, delete_pipeline, create_dashboard, delete_dashboard
# Environment variable
GREPTIMEDB_ALLOW_WRITE=true greptimedb-mcp-server
# Or CLI argument
greptimedb-mcp-server --allow-write true
When enabled, the security gate is bypassed for execute_sql, and the server
logs a warning on startup. With write mode off, those four tools return an error
without contacting the server; list_pipelines, list_dashboards and
dryrun_pipeline are read-only and stay available.
⚠️ Danger: This lets an AI assistant run destructive statements against your
database. Never enable it against production data. Combine with a read-only
database user if you only need read access.
Data Masking
Sensitive columns are automatically masked (******) based on column name patterns:
- Authentication:
password, secret, token, api_key, credential
- Financial:
credit_card, cvv, bank_account
- Personal:
ssn, id_card, passport
Configure with --mask-patterns phone,email to add custom patterns.
Audit Logging
All tool invocations are logged:
2025-12-10 10:30:45 - greptimedb_mcp_server.audit - INFO - [AUDIT] execute_sql | query="SELECT * FROM cpu LIMIT 10" | success=True | duration_ms=45.2
Disable with --audit-enabled false.
Development
# Clone and setup
git clone https://github.com/GreptimeTeam/greptimedb-mcp-server.git
cd greptimedb-mcp-server
uv venv && source .venv/bin/activate
uv sync
# Run tests
pytest
# Format & lint
uv run black .
uv run flake8 src
# Debug with MCP Inspector
npx @modelcontextprotocol/inspector uv --directory . run -m greptimedb_mcp_server.server
License
MIT License - see LICENSE.md.
Acknowledgement
Inspired by: