Skip to Content
Neo MCP

Neo MCP

Your coding agent, running AI/ML engineering tasks

You already use Cursor, Claude Code, or Copilot to edit code and answer questions. AI/ML work often needs more: run a training job, optimize or fix an AI agent, compare model evals, iterate on a pipeline — work that takes multiple steps, not one chat reply.

Neo MCP plugs Neo into that chat via the Model Context Protocol. You ask your agent; the agent starts a Neo task; Neo plans, runs, and debugs in your environment and writes code, metrics, and reports into your repo. You review changes in the IDE like any other work.

Neo MCP workflow: you ask your coding agent in the editor, the agent calls Neo over MCP, Neo runs the AI/ML task, and outputs land in your repository

You → coding agent → Neo (via MCP) → runnable artifacts in your repo

Set up Neo MCPWhat is Neo?

Works with Cursor, Claude Code, VS Code Copilot, Windsurf, Zed, Continue, Codex CLI, and other MCP editors.

Set up Neo MCP

1

Get your Neo API key

Create a Neo key in the dashboard. This connects your editor to Neo.

  1. Open API keys in the Neo dashboard (sign in if prompted).
  2. 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.

2

Install Neo MCP

Install the local MCP server with one Python command.

Python 3.11+:

pip install neo-mcp

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.

3

Add Neo to your editor

Choose your editor and paste the matching config with your API key.

Replace sk-v1-YOUR_KEY with your real key in the snippet for your editor.

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.

What you can ask Neo to do

Once Neo is connected, ask your coding agent in normal language. The agent will call the right MCP tool when Neo is useful.

  • Use Neo to fix this failing training run and save the results.
  • Use Neo to benchmark these prompts on our eval set.
  • Use Neo to build a RAG pipeline and write the runnable files into this repo.
  • Use Neo to fine-tune this model and summarize the eval metrics.

MCP tools

Neo exposes these tools to your editor. You usually do not call them manually; your coding agent uses them after you ask in plain English.

ToolWhat it does
neo_submit_taskStarts a Neo task and returns a thread_id.
neo_task_statusChecks whether the task is running, complete, paused, failed, or waiting for feedback.
neo_get_messagesReads the task transcript and final output.
neo_list_tasksLists recent or active tasks after reopening your editor.

Typical flow: submit → check status → send feedback if Neo asks → read messages.

Keys and access

There are three different kinds of keys on this page:

Neo API key

Connects your editor to Neo MCP. Required for setup.

Integrations

Lets a Neo task use external services like GitHub, HuggingFace, Anthropic, OpenRouter, or custom secrets.

BYOK

Lets Neo run on your own Anthropic, OpenAI, or OpenRouter model account.

Give Neo access to external services by telling your editor what to save:

  • “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.”

Supported built-in providers are github, huggingface, anthropic, and openrouter. You can also store a custom secret with your own provider name.

Troubleshooting

Most issues show up during setup. Here are the common ones and quick fixes.