mcp-parseable-server - A Parseable MCP Server

This project is currently in early development with focus on log data. Feedback and contributions are welcome.

Testing has been done using:

• vscode with Github Copilot as the agent.
• opencode agent CLI tool

[TOC]

Overview

This project provides an MCP (Message Context Protocol) server for Parseable, enabling AI agents and tools to interact with Parseable data streams (logs, metrics, traces) using natural language and structured tool calls.

Features

• List available data streams in Parseable
• Query data streams using SQL
• Get schema, stats, and info for any data stream
• Modular MCP tool registration for easy extension
• Supports both HTTP and stdio MCP modes
• Environment variable and flag-based configuration
• The mcp server returns responses in json where the payload is both in text and structured format.

In Parseable dataset and data stream names are used interchangeably, as Parseable's datasets are essentially named data streams. In all tool description we try to use the term data stream to avoid confusion with the term dataset which can have different meanings in other contexts.

Limitations

Authentication

The mcp server does not implement any authentication mechanisms. If this is needed, use a reverse proxy like nginx or envoy in front of the mcp server to add authentication and authorization.

MCP session management

The mcp server does not implement any session management since all tool calls are stateless. This may change in the future.

Building

Ensure you have Go 1.20+ installed.

git clone https://github.com/thenodon/mcp-parseable-server
cd mcp-parseable-server
# Build the MCP server binary
go build -o mcp-parseable-server ./cmd/mcp_parseable_server

Running

HTTP Mode (default)

./mcp-parseable-server --listen :9034

The MCP server will listen on http://localhost:9034/mcp for agent/tool requests.

Stdio Mode

./mcp-parseable-server --mode stdio

This mode is used for CLI or agent-to-agent workflows.

Configuration

You can configure the Parseable connection using environment variables or flags:

• PARSEABLE_URL or --parseable-url- url to the parseable instance (default: http://localhost:8000)
• PARSEABLE_USERNAME or --parseable-user (default: admin)
• PARSEABLE_PASSWORD or --parseable-pass (default: admin)
• LISTEN_ADDR or --listen - the address when running the mcp server in http mode (default: :9034)
• INSECURE - set to true to skip TLS verification (default: false)`
• LOG_LEVEL - set log level. Supported levels are debug, info, warn and error (default: info)

Example:

PARSEABLE_URL="http://your-parseable-host:8000" PARSEABLE_USER="admin" PARSEABLE_PASS="admin" ./mcp-parseable-server

Production deployment

For production deployment use a reverse proxy like nginx or envoy in front of the mcp server that manages authentication, authorization, and tls termination.

Testing

See TESTING.md for a complete testing guide.

MCP Tools Reference

1. query_data_stream

Execute a SQL query against a data stream.

• Inputs:
  • query: SQL query string
  • streamName: Name of the data stream
  • startTime: ISO 8601 start time (e.g. 2026-01-01T00:00:00+00:00)
  • endTime: ISO 8601 end time
• Returns: Query result and the row count returned

2. get_data_streams

List all available data streams in Parseable.

• Returns: Array of stream objects with count

3. get_data_stream_schema

Get the fields schema for a specific data stream.

• Inputs:
  • stream: Name of the data stream
• Returns: Schema fields and types

4. get_data_stream_stats

Get stats for a data stream.

• Inputs:
  • streamName: Name of the data stream
• Returns: Stats object (see tool description for details)

5. `get_data_str

... View full README on GitHub