Model Context Protocol (MCP) is Anthropic's open protocol for connecting AI assistants to external tools and data. GreptimeDB MCP Server v0.3 (latest: v0.3.1) adds data masking for secure AI queries, tools for managing log pipelines, and production-ready prompt templates with documentation references.

Breaking News: With the Linux Foundation establishing the Agentic AI Foundation, Anthropic has donated the MCP protocol to the foundation. Also donated are Block's Goose and OpenAI's AGENTS.md, among other projects.

If you're new to GreptimeDB MCP Server, check out our introduction blog post for background on how MCP bridges LLMs and observability databases.

What's New

Data Masking

When LLMs query databases, sensitive data can leak into responses. Ask Claude to query a users table, and you might see passwords or API keys in the output. v0.3 automatically masks credentials, financial data, and personal identifiers before they reach the LLM.

Built-in patterns cover three categories:

• Authentication: password, passwd, pwd, secret, token, api_key, apikey, access_key, private_key, credential, auth, authorization
• Financial: credit_card, creditcard, card_number, cardnumber, cvv, cvc, pin, bank_account, account_number, iban, swift
• Personal: ssn, social_security, id_card, idcard, passport

Matched values appear as ****** in all output formats. Configure via environment variables:

# Disable masking (enabled by default)
GREPTIMEDB_MASK_ENABLED=false

# Add custom patterns (comma-separated, extends defaults)
GREPTIMEDB_MASK_PATTERNS=phone,address,email

TQL and RANGE Query Tools

Three new tools for time-series analysis:

Tool	Description
execute_tql	Run PromQL-compatible queries with start, end, step, and optional lookback
query_range	Execute RANGE/ALIGN aggregations with time-window semantics
explain_query	Analyze SQL or TQL execution plans; use EXPLAIN ANALYZE for actual metrics

Example TQL query:

{
  "query": "rate(http_requests_total[5m])",
  "start": "2024-01-01T00:00:00Z",
  "end": "2024-01-01T01:00:00Z",
  "step": "1m"
}

For more on TQL syntax, see the TQL documentation.

Pipeline Management: AI-Powered Log Parsing

GreptimeDB's log pipeline transforms raw logs into structured data. v0.3 adds MCP tools to manage pipelines directly:

Tool	Description
create_pipeline	Create pipelines with YAML configuration
dryrun_pipeline	Test pipelines without writing to the database
delete_pipeline	Remove specific pipeline versions
list_pipelines	View existing pipelines and versions

The pipeline_creator prompt helps LLMs generate pipeline configs from log samples. It covers:

• Processors: dissect, regex, date, epoch, gsub, select
• Transform configuration and data types
• Index best practices: inverted, fulltext, skipping
• Table design for log data

See this example where Claude builds a pipeline from nginx logs, tests it with dryrun_pipeline, and deploys with create_pipeline.

For pipeline configuration details, refer to the Pipeline Configuration Reference.

Production-Ready Prompts

All seven prompt templates now include References sections linking to official docs:

Template	Use Case	Key References
pipeline_creator	Generate pipelines from log samples	Pipeline Config, Data Index
log_pipeline	Log analysis with full-text search	Full-Text Search, SQL Functions
metrics_analysis	Metrics monitoring	RANGE Query, Data Model
promql_analysis	PromQL/TQL guidance	TQL Reference
iot_monitoring	IoT device analysis	Table Design
trace_analysis	Distributed tracing (OpenTelemetry)	Traces Overview
table_operation	Schema diagnostics	INFORMATION_SCHEMA

We also fixed query syntax issues in the trace_analysis and table_operation templates.

Other Improvements

• HTTPS Support: Connect over TLS with GREPTIMEDB_HTTP_PROTOCOL=https or --http-protocol https
• Security Gate Fix: SHOW CREATE TABLE now works correctly
• Performance: HTTP connections reuse aiohttp.ClientSession
• Output Formats: All query tools support csv, json, and markdown

Getting Started

Install via pip:

pip install greptimedb-mcp-server

Or run with uv:

uv run -m greptimedb_mcp_server.server

Configure your MCP client (Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "greptimedb": {
      "command": "greptimedb-mcp-server",
      "args": [
        "--host", "localhost",
        "--port", "4002",
        "--database", "public"
      ]
    }
  }
}

For pipeline management, ensure the HTTP port (default 4000) is accessible.

What's Next

This release upgrades the MCP server to use the FastMCP API, currently with stdio transport. Next, we plan to support HTTP SSE transport, enabling the server to run persistently as a standalone service. See Issue #23 for details.

Learn More

• GitHub Repository - Source code, issues, and contribution guidelines
• PyPI Package - Installation and version history
• GreptimeDB Documentation - Full database documentation
• Slack Community - Questions and discussions