@hmldns/nautexPublic

Connectors#Claude Code #Cursor #Codex #Gemini CLI#Coding #Product #Testing

MCP server for Nautex planning in coding agents

Nautex connects a coding agent to a project’s requirements, file map, and implementation plan through MCP. The setup flow writes agent config and rule files into the project so the agent can pull the next scope and follow the plan step by step.

85 stars•8 forks•Python•Updated 16d ago

Who it's for

Builders who want their coding agent to work from a project plan, file map, and synced requirements.

What it delivers

You can keep an agent aligned with the right files, tasks, and validation steps instead of re-explaining the project each time.

What it does

Requirements sync

Pulls PRD and TRD guidance from Nautex into the project for the agent to use.

File map and task flow

Links requirements to files and breaks work into coding, testing, and review tasks.

Agent-specific setup

Writes MCP config and rules for Claude Code, Cursor, Codex, OpenCode, Gemini CLI, and Grok Build.

Local project state

Stores project config in `.nautex/config.json` and token data in `.nautex/.env`.

Setup and probe scripts

Includes scripts to set up sessions and probe agent integrations.

How to get it

1The fastest way to set up is via the web app onboarding flow, which generates a single…
```
uvx nautex setup --token <TOKEN> --project <PROJECT_ID> --plan <PLAN_ID> --agent <AGENT>
```
2Alternatively, run the interactive terminal UI
```
uvx nautex setup
```

3On macOS and Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

4On Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

README

This is an MCP server that integrates PRD and TRD building tool Nautex AI with the Coding Agents.

Supported agents:

Claude Code
Cursor
Codex
OpenCode
Gemini CLI
Grok Build

Motivation

Since LLM Coding Agents do not attend team meetings, there is the challenge of conveying complete and detailed product and technical requirements to them.

Nautex AI tool-chain manages step by step guiding of Coding Agents so they implement specification using small, relevant and testable steps.

Core principles are:

start from foundational parts, de-risk them, then build up;
do not overwhelm Coding Agents by large problem at once;
plan project files map and link them to requirements and to tasks: Coding Agents don't get lost, you know how to navigate brand new code base;
manage developer attention for verification and validation in right moment for review.

How It Works

Nautex AI acts as an Architect, Technical Product Manager, and Project Manager for coding agents, speeding up AI-assisted development by communicating requirements effectively. This MCP server pulls guidance instructions from Nautex AI; tasks contain to-do items, references to the affected files, and requirements that are automatically synced for the Coding Agent's availability.

By Ivan Makarov

Usage Flow Presentation (unfold me)

Requirements Specifications

The chatbot conducts a briefing session with you, gathering questions and ideas until complete. It then generates comprehensive product and technical specifications.

(Example: A project I initiated to explore WebRTC.)

Product requirements: howitworks_specifications

Technical requirements:

Specification Refinement

You fill in details, clarify the specification, and resolve any TODOs flagged by the chatbot during the interview.

howitworks_refinement

Codebase Map and Project Files

You'll occasionally need to review the code, so it's best to know in advance where to look and how everything is organized. This prevents the AI from making decisions—allowing it to focus on writing higher-quality code with greater attention to the task.

The image displays a file map generated by Nautex AI, with files linked to specific requirements and sections.

howitworks_filemap

Agent Tasks

With the code location clarified, tasks are planned: Coding, Testing, and Review.

Reviews are scheduled early to demonstrate progress and verify alignment with goals.

The plan is structured in small, self-contained layers, building your project incrementally like floors in a skyscraper.

howitworks_tasks

Integration

Configure the MCP server for your coding agent: connect to the Nautex cloud platform, select the project, and choose the implementation plan. The setup command writes all configuration to your project root.

howitworks_integration

Coding with Coding Agents

In agent mode, instruct: "pull nautex rules, and proceed with the next scope."

At this stage, your specifications are synchronized in the .nautex directory and accessible to the Coding Agent. The MCP server continuously monitors their relevance.

That's it. You then review and accept substantial code segments that fully align with your expectations and requirements.

howitworks_coding

Setup

Quick Setup (one command)

The fastest way to set up is via the web app onboarding flow, which generates a single command you copy and run in your project root:

uvx nautex setup --token <TOKEN> --project <PROJECT_ID> --plan <PLAN_ID> --agent <AGENT>

Parameters:

Flag	Description
`--token`, `-t`	API token (create at nautex.ai)
`--project`, `-p`	Project ID
`--plan`, `-l`	Implementation plan ID
`--agent`, `-a`	Agent type: `claude`, `cursor`, `codex`, `opencode`, `gemini`, `grok`
`--yes`, `-y`	Skip confirmation prompts

This validates your token, project, and plan, then writes all configuration to your project root:

.nautex/config.json — project config
.nautex/.env — API token (git-ignored)
MCP config — agent-specific (see below)
Agent rules — merged into existing rule files without overriding your content

Interactive Setup (Terminal UI)

Alternatively, run the interactive terminal UI:

uvx nautex setup

Setup Screenshot

Follow the on-screen prompts to select your project, plan, and agent.

How to Install UV

On macOS and Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

On Windows:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Check the latest instructions from the UV repo for details and updates.

What Gets Written Per Agent

All configuration is scoped per-project in your project root.

Claude Code

MCP: registered via claude mcp add nautex -s local -- uvx nautex mcp
Rules: managed section added to CLAUDE.md
Verify: run claude mcp list and check for nautex: uvx nautex mcp

Cursor

MCP config: .cursor/mcp.json
Rules: .cursor/rules/nautex_workflow.mdc

Note: After setup, Cursor may ask via popup whether to enable the new MCP — answer yes. In any case, go to File -> Preferences -> Cursor Settings -> Tools & Integrations and make sure the Nautex MCP toggle is enabled (green).

Codex

MCP config: .codex/config.toml (project-local, backup created as config.toml.bak before first write)
Rules: managed section added to AGENTS.md
Verify: use the /mcp command inside Codex to confirm nautex is listed

OpenCode

MCP config: opencode.json (project root, preserves unrelated fields, backup as opencode.json.bak if unparsable)
Rules: managed section added to AGENTS.md
Verify: invoke the Nautex MCP tool from OpenCode and run status

Gemini CLI

MCP config: .gemini/settings.json
Rules: managed section added to GEMINI.md

Grok Build

MCP config: .grok/config.toml (project-local, backup created as config.toml.bak before first write; startup_timeout_sec = 20)
Rules: managed section added to AGENTS.md
Verify: run grok mcp list / grok mcp doctor nautex, or open /mcps in the Grok TUI

Start Coding

Once setup is complete, launch your coding agent and tell it:

Check nautex status

After confirming the connection works:

Pull nautex rules and proceed to the next scope

Proceed with the plan by reviewing progress and supporting the Agent with validation feedback and inputs.

Prerequisites

Before running setup, prepare your project in the Nautex web app:

Sign up and create an API token
Create PRD and TRD documents (chat with the bot to capture requirements)
Create a files map of the project
Create an implementation plan

The web app onboarding flow will generate the setup command with all IDs pre-filled.

Projects built with nautex

Collaborative Pixel Canvas - repo

Best practice from the community

Files in the repo

Repository payload•13 top-level entries

doc— Diagrams and screenshots explaining the workflow and setup.
scripts— Setup, session, and probe scripts for agent and MCP configuration.
src— Source code
tests— Tests
tools— 1 item
.gitignore
AGENTS.md— Project rules file used by agent workflows.
CLAUDE.md— Claude Code rules file used during setup.
LICENSE— License
Makefile
pyproject.toml— Python project metadata and packaging.
README.md— The README
requirements.txt— Runtime dependencies.

Discussion (0)

Ask about usage, or say what you built with it

No comments yet. Be the first to say what this is good for.

More connectors

koala73/worldmonitorConnectors

Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface

86k

DeusData/codebase-memory-mcpConnectors

High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 158 languages, sub-ms queries, 99% fewer tokens. Single static binary, zero dependencies.

43k

lidge-jun/opencodexConnectors

Universal provider proxy for OpenAI Codex & Claude Code — use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama…) with Codex CLI, App, SDK, and Claude Code

14k

okf-memory/

okf-agent-memory

okf-memory/okf-agent-memoryConnectors

Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zero external databases or dependencies. Built in pure Go.

547

tirth8205/

code-review-graph

tirth8205/code-review-graphConnectors

Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.

31k

2akouwu/

reverify

2akouwu/reverifyConnectors

Stop your AI from making things up — it proposes, deterministic tools decide, every claim checked against ground truth with evidence. Grounded facts and context survive resets. Reverse engineering is the proving ground. MCP server + CLI.

1.1k