Take the tour
Click a step to see that part of the app.
Claims, experiments, evidence and related work live as plain Markdown. You see an argument map in VS Code; your agents read and write the same records through MCP — and you keep the final say on every verdict.
Every screen on this page uses the bundled demo project with synthetic data.
Click a step to see that part of the app.
One memory folder, three ways in: the VS Code app, the research CLI, and an MCP server for agents.
Each record is a Markdown file with YAML front matter. A local index keeps it fast, and every change keeps the previous version in history/.
A project handoff is about 20 KB instead of megabytes. Agents read one claim's evidence at a time, search ranked cards, and check their records with research_lint.
Agents propose verdicts and back-burner moves; only you accept them. Nothing is deleted: refuted work keeps its evidence and can be restored.
About five minutes, with the demo project.
Download research-manager-0.1.9.vsix from Releases, then in VS Code choose Extensions → … → Install from VSIX and run Research Manager: Open.
Agents use the research command from a clone of the repository. On Linux the bootstrap installs Node 24 locally and builds; on macOS or Windows install Node 24, then run npm ci && npm run build.
git clone https://github.com/HanpuShen/research-manager.git cd research-manager && ./scripts/bootstrap.sh
To see the demo, run Research Manager: Set Memory Folder and choose examples/demo-memory in the clone.
Any MCP client works. With Claude Code, from the clone:
claude mcp add research-manager -- "$PWD/bin/research" \ --memory /path/to/research-memory mcp
Then ask: “Read the handoff for my project and tell me what needs a decision.” Add --read-only for an agent that should analyse but not write.