Nauro

Community

Give your agents the context code leaves out.

★ 10 Stars Category · Dev Tools Popular Source revision ad549d7c0d69

58FMRS · C

Nauro is a local-first MCP server that provides a shared project memory for AI coding agents, emphasizing human approval and decision rationale. Its install configuration is via uvx/PyPI, and it is suitable for long-lived projects. Official status is not claimed, but the server is maintained by the upstream project's maintainers.

Reliability

10/20

Security and permissions

12/20

Maintenance

12/20

Documentation

12/20

Setup experience

12/20

Read the FMRS scoring method →

Nauro keeps current state, open questions, and human-approved project judgment in one record, ready for every agent you connect. The record combines project scope, state, open questions, and the rationale behind decisions. Nauro surfaces the relevant slice before work, then carries approved judgment and reported progress into later sessions and connected tools.

Tools

check-decision

Submit a new decision to the project record for checking and approval.

Setup

Install Nauro with uv tool install nauro. Ensure uv is installed first (macOS/Linux: curl -LsSf https://astral.sh/uv/install.sh | sh; Windows instructions in docs). Alternatively, with Python 3.10+, run pipx install nauro. After installation, cd into your repository and run nauro adopt --with-skills --with-subagents, restart your agent, then invoke /nauro-adopt in Claude Code or $nauro-adopt in Codex.

Fit and risk

Best for

Long-term projects where project context decays across sessions, tools, and handoffs
Pairing with AI coding agents like Claude Code, Codex, or Cursor to reduce redundant questions and context loss
Teams that value local data ownership and human-approved decision records

Not for

Small repositories: if a reliable AGENTS.md or CLAUDE.md keeps agents oriented, Nauro may be overkill
CI/CD pipelines requiring strict enforcement on every change—Nauro's retrieval is advisory and never blocks
Users wanting fully automatic decision logging without human approval gates

Required permissions

The local record is plain Markdown that you own; local use needs no account and sends no analytics
Nauro can read project files (e.g., README, code) to assess decision rationale but writes only after your explicit approval
Installation may add skill files and workflow agents (e.g., nauro-ship-task) if requested
Explicit overwrite via `nauro sync` is the only controlled override path, preserving manually managed files

Risks and side effects

Unmanaged AGENTS.md may be overwritten—Nauro preserves it during adoption; `nauro sync` is the sole explicit overwrite path with a `# Manual` section protecting it
As an MCP server, it uses stdio; any client with access to the MCP can read the local record content
Nauro does not enforce decisions on agents; it is advisory, so agents might ignore stored judgment

Troubleshooting

Run `nauro status` to verify MCP, skills, and workflow agents are properly installed
Run `nauro doctor` to check the health of the project store
If skills are not loading, ensure you adopted with `--with-skills --with-subagents` and restart your agent
For sync issues, check if `nauro sync` is unavailable or if files are in a manual state

Use cases

Maintain a shared project memory across sessions, tools, repositories, or machines in longer-lived work

Quickly bring new agents or sessions up to speed on project state and decision rationale

Record and build consensus around key architectural decisions, e.g., storing dollar amounts in decimal

Supported clients

Claude DesktopPartial support

Related MCP servers

Compare with other MCP servers