Model Context Protocol (MCP) - Private Preview

Note

This is a private preview feature. Please contact your Observe representative for more information or open Docs & SupportContact Support in the product and let us know.

Model Context Protocol (MCP) is an open standard for sharing context between AI models and data systems.
The Observe MCP Server lets agentic tools such as Augment, Cursor, Claude Code, and Claude Desktop connect securely to your Observe tenant so they can investigate incidents, or explore logs, metrics or traces.

What data leaves Observe?

Observe MCP Server

Event

What we send to OpenAI

Purpose

Knowledge graph build / refresh
(runs periodically)

• Dataset schema (names & types)
≤ 10 sample rows per dataset
• Column‑level stats
• Metric names & discovered tag keys/values

Allows the agent to reason over your dataset catalogue.

Runtime query
(each time an agent executes an query)

• A short dataset summary document for the candidate tables (no raw rows)

Helps the model choose the correct dataset(s) to answer the user’s question.

OpenAI doesn’t train their models on your data by default, but we’d like to be transparent about the data that is sent to OpenAI when you use our MCP server.

Configure the Observe MCP Server

  1. Go to Workspace settings page and find MCP Server details page.

  2. Pick your tool of choice.

MCP Server details
  1. Configure the tool to connect to the Observe MCP Server.

  2. Restart your tool.

  3. Start asking a quetion such as “What are the top 5 errors in my application? Use the observe mcp server to answer the question.” to your agent.