nauro 0.1.0

Local CLI + MCP server that maintains versioned project context for AI coding agents.

Nauro

Persistent project context for AI coding agents.

Nauro maintains versioned project context (decisions, state, open questions) and delivers it to Claude, Perplexity, ChatGPT, Cursor, and any MCP client. Think git log for why your project is the way it is.

Install

pipx install nauro

Or with pip:

pip install nauro

Requires Python 3.11+.

Quickstart

Try the demo locally (1 minute)

nauro init --demo
nauro setup claude-code

Open Claude Code and ask:

"What did we decide about the database?"

The demo creates a sample project with three decisions, project state, and open questions. No account needed.

Try it across surfaces (2 minutes)

To access the same project context from Claude AI, Perplexity, or ChatGPT:

nauro auth login
nauro sync

Then add mcp.nauro.ai as a remote MCP connector in your tool's settings. Ask the same question — same answers, different surface.

Use with your project

# Register your project
nauro init my-project

# Log a decision directly
nauro note "Chose Postgres over MongoDB for ACID compliance"

# Set your Anthropic API key to auto-extract decisions from commits
nauro config set api_key sk-ant-...
nauro extract

Why Nauro?

Approach	Cross-tool	Extracted from commits	Versioned	Format
Nauro	All MCP clients	Yes (via Haiku)	Snapshots + diffs	Portable markdown
AGENTS.md (manual)	Tools with repo access	No	Git history only	Markdown
Cursor Rules	Cursor only	No	No	Proprietary
Claude Memory	Claude only	Partial	No	Proprietary

check_decision catches when a new approach conflicts with a past decision, across any connected surface.

How it works

A Python CLI extracts decisions from your git history using Haiku, stores them as flat markdown in ~/.nauro/projects/, and validates new decisions against existing ones (structural screening, embedding similarity, LLM evaluation). An MCP server delivers context to any connected AI tool. Cloud sync keeps everything in sync via S3.

~/.nauro/projects/<n>/
  project.md          # goals, constraints
  state.md            # current focus, blockers
  decisions/          # one markdown file per decision
  open-questions.md   # unresolved threads
  snapshots/          # versioned store captures

All content is plain markdown. No database, no proprietary format.

MCP tools

11 tools (7 read, 4 write) exposed to any connected MCP client:

Read:

• get_context — project summary at three detail levels (L0/L1/L2)
• list_decisions — browse the full decision history
• get_decision — full content of a specific decision by number
• search_decisions — keyword search across decision titles and rationale
• get_raw_file — raw markdown content of any store file
• diff_since_last_session — what changed since your last session (or N days ago)
• check_decision — check a proposed approach for conflicts without writing

Write:

• propose_decision / confirm_decision — write decisions with conflict validation
• flag_question — flag an unresolved question
• update_state — report progress

Your data

Local extraction (free tier): Code diffs go directly from your machine to your Anthropic API key. Nauro is never in the data path.

Cloud sync: Project context (decisions, state, open questions — not source code) is stored encrypted in AWS S3 (SSE-S3). Each user's data is isolated under a unique prefix.

Remote MCP: When connected, your project context is read from S3 and delivered to the AI tool. The AI tool's own data policies apply after delivery.

Pricing

Free tier: unlimited local usage + 100 remote MCP calls/month. Pro ($9/mo) adds unlimited remote MCP and hosted extraction.

Contributing

Contributions welcome. See the project structure in CLAUDE.md for architecture context.

pip install -e ".[dev]"
pytest tests/ -x -q -m "not integration"

870+ tests across nauro, nauro-core, and mcp-server. Apache 2.0 license.