Neo-mcp
Neo is the first autonomous AI engineering agent that plans, researches and executes multi-step work for tasks such as building AI agents, AI model fine-tuning and evals, and ML pipelines; using your codebase, data, and experiments to ship faster with inspectable artifacts. It can reason over your repository, wire tools and retrieval, debug training runs, and help you develop production-ready AI workflows
CTRL K
- Getting Started
- Meet Neo
- Quick Start
- Core Concepts
- VS Code Extension
- Cursor IDE
- Neo MCP
- Neo for Claude Code
- Bring Your Own LLM Keys
- Learn More
- Projects
- Production-Ready A/B/n Testing Framework
- Adversarial Robustness Probe: Stress-Testing NLP and Vision Models
- Natural Language Time Series Forecasting with TimesFM
- Multilingual ASR Pipeline with Qwen3-ASR-0.6B
- Attention Head Visualiser: Mapping GPT-2 Heads
- AutoDoc: Autonomous Documentation Agent
- AutoPrompter: Closed-Loop Prompt Optimization
- Carbon-Aware Model Training Pipeline
- Code Agent Analysis and Refactoring Tool
- Low-Latency CPU-Based Voice Assistant with Streaming TTS
- Embedding Evaluator: Vector Quality & Drift Audits
- GitHub Repo Query Agent
- GPU Scout: Cloud GPU Monitoring & Spot Alerts
- Hallucination Benchmark Tool
- Helix: Self-Healing CI/CD Agent
- Medical Report Analysis (HIPAA)
- How to Fine-Tune an LLM Using NEO
- Hypothesis Engine: Autonomous Research Framework
- Invoice OCR Pipeline
- Latest LLMs: Real-Life Task Evaluation
- LLM Consistency Monitor: Detect Prompt Brittleness
- LLM Council: Multi-Model Deliberation
- LLM Evaluator Tool
- LLM Response Judge: Auto-Grade AI Responses
- MCP Optimization & Suggestion Agent
- Multi-Agent Memory Management System
- Multi-Agent Observability Platform
- Multi-Model RAG System
- Multi-Query Batch Inference Optimization
- NeuroLens: Concept Neuron Discovery
- Orbit: 3D Dependency Visualiser
- Prompt A/B Testing Tool: Statistical LLM Evaluation
- Prompt Injection Defence System
- Quantization-Aware Training for Edge Deployment
- Real-Time Voice Translation
- Spec to Ship: Multi-Agent Feature Delivery
- Stock Trading Agent Swarm
- Table Extraction from Financial Documents
- Platform Features
- Use Cases
- FAQ
- Resources
- Dashboard ↗
- Datasets
- Settings
Light
Neo MCP
Use the Model Context Protocol (MCP) to wire your editor to Neo: the neo-mcp package runs the server locally and syncs results into your project folder.
Get your API key
- Open API keys in the Neo dashboard (sign in if prompted).
- Create a key and copy it. Keys look like
sk-v1-…— treat them like secrets and do not commit them to git.
You will paste this value wherever the examples use sk-v1-YOUR_KEY.
Install
Python 3.11+:
pip install neo-mcp
LinuxWindows
If pip install neo-mcp fails with error: externally-managed-environment, run:
python3 -m pip install --user --break-system-packages neo-mcp
Installs to ~/.local/ without sudo.
Editor configuration
Replace sk-v1-YOUR_KEY with your real key in the snippet for your editor.
Claude CodeCursorWindsurfVS CodeZedContinueCodex CLIAntigravity
Claude Code configures MCP through the claude mcp CLI (run these in your system terminal, not inside a chat).
1. Register Neo
claude mcp add --scope user neo \
-e NEO_SECRET_KEY=sk-v1-YOUR_KEY \
-- python3 -m neo_mcp
2. Confirm it’s connected
claude mcp list
You should see neo with a green checkmark.
Demo: Run a task through MCP
MCP tools
| Tool | What it does |
|---|---|
| neo_submit_task | Start a Neo task; returns a thread_id. Optional wait_for_completion blocks until finished. |
| neo_task_status | Poll status (RUNNING, COMPLETED, WAITING_FOR_FEEDBACK, etc.). |
| neo_get_messages | Read the task transcript after completion (large output may be capped). |
| neo_list_tasks | List recent or active tasks (e.g. after reopening the editor). |
| neo_send_feedback | Reply when Neo is waiting for input (WAITING_FOR_FEEDBACK). |
| neo_pause_task / neo_resume_task | Pause or resume a task. |
| neo_stop_task | Stop and tear down a task. |
| neo_add_integration | Store a provider key (GitHub, HuggingFace, Anthropic, OpenRouter, or any custom secret) on your machine so Neo can use it in tasks. |
| neo_list_integrations | List configured integrations (provider names only — never prints the secret). |
| neo_test_integration | Call the provider’s API once to confirm the stored key is still valid. |
| neo_remove_integration | Delete a stored key and its file (or keyring entry). |
Typical flow: submit → poll status until COMPLETED or feedback → get messages (or use wait_for_completion on submit to skip polling).
Integrations
Give Neo access to external services — GitHub, HuggingFace, Anthropic, OpenRouter, or any custom secret — by storing API keys on your machine once. Neo injects them as environment variables every time it runs a task, so your scripts, CLIs, and libraries just work. Keys never leave your machine.
Adding a key
Just tell Neo in plain English — from any MCP-connected editor:
“Add my Anthropic key
sk-ant-...to Neo.”
“Store my GitHub PAT
ghp...so Neo can push to my repos.”
“Save my HuggingFace token
hf...for downloading models.”
Neo calls neo_add_integration behind the scenes. You only need to do this once per provider — the key is stored locally and reused across every future task.
| Provider | Credential | What Neo can do with it |
|---|---|---|
| github | Personal Access Token (ghp_...) | Clone private repos, push, open PRs |
| huggingface | Token (hf_...) | Download private models and datasets |
| anthropic | API key (sk-ant-...) | Run Claude models inside your tasks |
| openrouter | API key (sk-or-...) | Route through any model OpenRouter supports |
You can also store any custom secret the same way — use any provider name and Neo will save it and expose it to tasks.
Where keys are stored
File on disk (default)OS keyring (safer)
Keys are saved to a small file on your machine at ~/.neo/integrations/<provider>.env. The file is owner-read-only — no other user on your computer can access it. This is the same way tools like gh, aws, and huggingface-cli store their tokens.
For GitHub and HuggingFace, Neo also writes the token where the official CLI expects it (~/.git-credentials and ~/.cache/huggingface/token), so commands like git clone and huggingface-cli just work on their own — without Neo running.
- Works everywhere: Mac, Linux, Docker, servers, CI.
- Nothing extra to install.
Tip: turn on full-disk encryption (FileVault on macOS, BitLocker on Windows, LUKS on Linux) for an extra layer of protection when your machine is powered off or stolen.
How Neo uses your keys
When Neo runs a task, it reads your configured keys and sets them as environment variables on each child process:
| Variable | Provider |
|---|---|
| ANTHROPIC_API_KEY | anthropic |
| OPENROUTER_API_KEY | openrouter |
| HF_TOKEN / HUGGING_FACE_HUB_TOKEN | huggingface |
| GITHUB_TOKEN / GH_TOKEN | github |
Any script, CLI, or library that reads these env vars works out of the box — no per-task configuration needed.
Managing integrations
neo_list_integrations # shows configured providers (names only, never the key)
neo_test_integration anthropic # calls the provider's API to confirm the key is valid
neo_remove_integration anthropic # deletes the stored key and its file or keyring entry
Security
- Keys never leave your machine — they are never sent to the Neo backend, written to logs, or committed to git.
~/.neo/and.envfiles are already in.gitignore.- Secret files are always mode
0o600. neo_list_integrationsreturns provider names only, never the secret itself.- Use keys with the smallest scope you can, and rotate them if you suspect anything went wrong.
関連サーバー
Alpha Vantage MCP Server
スポンサーAccess financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
Pinelabs MCP Server
The Pine Labs Online MCP Server implements the Model Context Protocol (MCP) to enable seamless integration between Pine Labs’ online payment APIs and AI tools. It allows AI assistants to perform Pine Labs Online API operations, empowering developers to build intelligent, AI-driven payment applications with ease.
Layered Code
An AI-assisted web development tool for creating, modifying, and deploying code through natural language conversations.
Taeks MCP Server
An example MCP server designed for deployment on Cloudflare Workers, operating without authentication.
Yellhorn MCP
An MCP server that integrates Gemini 2.5 Pro and OpenAI models for software development tasks, allowing the use of your entire codebase as context.
WatchTower
Cost tracking + security scanning for AI builders
Roblox Studio MCP Server
Provides AI assistants with comprehensive access to Roblox Studio projects for exploration, script analysis, debugging, and bulk editing.
Adamik MCP Server
Interact with over 60 blockchain networks using any MCP client. Requires an Adamik API key.
Bellwether
Open-source CLI for testing MCP servers and detecting schema drift
Remote MCP Server (Authless)
An example of a remote MCP server deployable on Cloudflare Workers without authentication.
eBPF MCP
A secure MCP server for eBPF, designed for AI integration, kernel introspection, and automation.