How Doen Works

Doen collects context over time from your product engineering work—documents, issues, and discussions—and makes it available to AI agents through the Model Context Protocol.

AI coding assistants are powerful, but they're only as good as the context they receive. When you ask Claude Code to implement a feature, it needs to understand not just the code, but the product requirements, past decisions, and team conventions that shape how that feature should be built.

Doen is designed to solve this problem. It collects context from your product engineering work over time, organizes it into a searchable knowledge graph, and makes it available to AI agents through the Model Context Protocol (MCP).

The Three Pillars of Context

Doen structures product context around three core entity types, each serving a distinct purpose in capturing and organizing knowledge:

Documents

Documents capture the "why" and "what" of your product. They include:

• Product specifications: Feature requirements, user stories, acceptance criteria
• Architecture notes: Technical decisions, system design, API contracts
• Decision records: Why the team chose a particular approach over alternatives
• Custom documentation: Any long-form content relevant to your product

Documents are versioned, so the AI can understand not just the current state, but how requirements evolved over time. When you update a spec, the previous version remains accessible for historical context.

Issues

Issues represent units of work—features to build, bugs to fix, improvements to make. Each issue captures:

• Title and description: What needs to be done and why
• State: Open, in progress, in review, done, or canceled
• Assignments: Who is responsible for the work
• Scope/labels: Categorization for filtering and organization
• Parent/child relationships: Hierarchical breakdown of complex work
• Activity history: What happened, when, and who did it

Comments and Discussions

Attached to issues, comments capture the informal dialogue that shapes implementation. Questions, clarifications, alternatives considered, and decisions made—all become part of the permanent context. Unlike Slack threads that disappear into the void, Doen comments remain attached to the work they describe, creating a discoverable history of team thinking.

How Context Accumulates

Doen doesn't require you to manually "add context." Instead, context accumulates naturally as you work:

Creating and Updating

Every document you write, every issue you create, every comment you post becomes part of the context graph. The act of doing your normal work builds the knowledge base.

Linking

References between entities create relationships. When you mention issue #42 in a document, that creates a bidirectional link. The document knows about the issue, and the issue knows about the document. These links are explicit and navigable. An AI agent working on issue #42 can automatically retrieve the related documents that define what should be built.

Integration with Development Tools

Through webhooks, Doen connects to your development workflow. When commits reference issue numbers, Doen links them: push events show what code changed, pull requests link to the issues they address, and commit messages with issue references create automatic associations.

The MCP Interface

Doen exposes context through the Model Context Protocol, making it accessible to AI tools like Claude Code. The MCP server provides:

• Semantic Search: AI agents can search across issues and documents using natural language
• Structured Retrieval: Agents can retrieve specific issues by number, fetch documents by type, and navigate the relationship graph
• Issue Operations: Beyond reading, the MCP interface allows agents to update issues—changing state, adding comments, logging activity
• Project Overview: Agents can request a high-level summary of the project

Why This Approach Works

• Low Friction: Context collection happens through normal work activities
• Always Current: Because context comes from active work, it's automatically up to date
• Discoverable: Semantic search and explicit linking make information findable
• Structured for AI: Unlike chat histories or wiki pages, Doen's context is structured specifically for AI consumption

Conclusion

Doen's approach to context is simple in concept: capture the artifacts of product engineering work, link them together, and make them accessible to AI. The power comes from consistency and accumulation—every document, issue, and comment becomes part of a growing knowledge graph. For teams using AI-assisted development, this context is invaluable. Instead of manually explaining context each session, the AI has access to the full history of your product.