Give agents the context
to do useful work.

The problem

An agent’s usefulness is limited when the context it needs is scattered across tickets, documentation, source code and build systems. Manually gathering that information becomes another task for the engineer.

I built RAVEN to give agents a consistent way to work with those systems through the Model Context Protocol, or MCP.

What I built

The platform includes integrations for Jira, Confluence, Bitbucket, GitHub, Azure DevOps, Jenkins and other engineering tools. It exposes purpose-built tools for actions such as finding related issues, reading documentation, examining code and checking builds.

I worked on the tool interfaces, authentication, personal-information scrubbing, tests and release tooling. I also built workflows on top of those integrations, including fault investigation and fix preparation.

1. 01Give the agent a defined engineering task
2. 02Retrieve relevant context through MCP tools
3. 03Use authenticated, scoped integrations
4. 04Inspect the evidence and proposed changes

Making the tools dependable

Connecting an API is one part of the work. The tools also need clear inputs, useful responses, predictable failures and boundaries on what they can do.

• Access: authentication and controls appropriate to each connected system.
• Privacy: personal-information scrubbing before model-facing output.
• Verification: automated tests, a generated tool inventory and regression coverage for security fixes.
• Usability: setup documentation and tool definitions that help both developers and agents understand the available operations.

The result

Colleagues use and contribute to RAVEN. It supports engineering investigation and provides the integrations used by an autonomous DevOps workflow. The source and documentation are public, so its implementation can be inspected directly.