Contextor is a context store for AI agents. It helps teams draft specs, PRDs, plans, and conventions with AI assistance, store them once, and let agents retrieve relevant context automatically via MCP—no more manual copy/paste of project knowledge.
This repository is a Bun-powered monorepo. The primary application lives in packages/app, and supporting assets (documentation, MCP tooling experiments, etc.) live alongside it.
# Install workspace dependencies
bun installbun run dev– start the Contextor app locally (Vite dev server)bun run build– build all workspaces (use--filter <workspace>for targeted builds)bun run check-types– run TypeScript project references in no-emit modebun run lint– run Biome with autofix enabledbun run test– execute Vitest against the app workspace (TypeScript targets only)bun run release– build the MCP server workspace and publish with Changesets (CI usually handles this)
Vitest currently focuses on non-React TypeScript modules (Convex helpers, utilities, etc.). Extend coverage as you add critical logic.
packages/app– Frontend/Convex implementation of the context storedocs– Product plans, specifications, and project overview docspackages/mcp-server– MCP integration tooling (experiments)
Each workspace manages its own scripts; use bun run --filter <workspace> <script> for workspace-specific tasks.
- Create a feature branch.
- Make changes scoped to a single concern.
- Run
bun run check-types,bun run lint, and (when applicable)bun run test. - Submit a PR with screenshots of UI changes and notes about any backend/schema updates.
Refer to AGENTS.md for detailed conventions around project structure, naming, and agent responsibilities.