Guide

Coding agents

Your coding agents report to Neurova Studio on your own Mac: what they work on, for which client, how long and at what cost. You can answer their questions from your iPhone or Watch.

On this page

Connect your agents

Neurova Studio follows Claude Code, OpenAI Codex, Cursor, GitHub Copilot CLI and Gemini CLI, through their hooks.

  1. Open Settings → Coding agents.
  2. Click Connect next to each agent you use. Studio adds its hook to that agent’s settings.
  3. For each project, add its repository folder on this Mac: on the project page, or under Folders on this Mac with Add folder…. Sessions started in that folder count for that project and its client.

Claude Code reports straight to Studio. Codex, Cursor, Copilot CLI and Gemini CLI get a one-line command that gives up after two seconds, so a closed Studio never slows them down. Your own hooks, and Neurova Awake’s, stay as they are.

Agents report only to Studio on this Mac (127.0.0.1), protected by a token in your Keychain. Nothing is sent anywhere else; your team sees the sessions through the encrypted sync. Studio must be open to receive them.

The Agents section

Agents in the sidebar shows:

  • Now: the sessions running at this moment, with their active time and number of tool calls;
  • Waiting for approval: an agent asking for permission, with Allow and Deny;
  • Recent sessions, and Agent work per client for the month.

Time and cost per client

Each session counts its active time (only while the agent is working) and, for Claude Code, its tokens, read from Claude Code’s own files in the ~/.claude folder you allowed when connecting it. The cost is an estimate at Anthropic’s API list prices. If you pay a flat subscription, your real cost differs, but the estimate is a fair measure to bill a client by.

Approve from iPhone and Watch

When Claude Code asks for permission, Studio can ask your other devices first:

  1. In Settings → Coding agents, under Approve from iPhone and Watch, switch on When Claude Code asks for permission, ask on my other devices first.
  2. Choose how long to Wait for an answer: 1, 2, 5 or 9 minutes.

Turn it on when you step away. The question goes to your agency’s devices as an encrypted record; you answer with Allow or Deny on your iPhone, Watch, another Mac or the web app. If nobody answers in time, Claude Code asks at the terminal as usual. Nothing is ever allowed without a person.

Notices on your devices are made there after a sync; no push service carries the question, so its text never passes through Apple’s servers.

Hand a request to an agent

A request linked to a project with a repository folder can go to an agent in one click:

  1. Open the request. Under Hand to a coding agent, choose the Agent and the Folder.
  2. Click Start in Terminal.

Studio writes a brief from the request and its notes (with your team’s handoff guidelines), makes a new branch in its own git worktree inside the repository, and starts your agent there in Terminal. The session’s time and cost, its pull request and a summary come back on the request.

The MCP server

Studio has a local MCP server, so an agent can read and update your records while it works: what needs attention today, clients, projects, requests (read, change status, priority or assignee), notes, tasks (list, create, update), the timer and agent sessions. It never sends anything to clients.

In Settings → Coding agents, under MCP server, Studio shows a command to run once in a terminal to add it to Claude Code. The server listens only on your Mac and needs its token; the tools see only what your role may see. New token… disconnects every agent and MCP client until you connect them again.

Workspace kit

Each project page has a Workspace:

  • Environments: each with its name and address.
  • Secrets: API keys and passwords for the project, kept in this Mac’s Keychain. They never sync and never reach agents through Studio. Copy clears the clipboard after a minute; Copy all as .env copies them in that format.
  • Write CLAUDE.md and Write AGENTS.md: your team’s templates, filled in for this project, written into its repository folder. If the repository already has one, Studio asks first; git keeps the old version if it was committed.
  • A handover document: Markdown with environments, hosting, open work and effort. Secrets are listed by name only.

Templates are under Settings → Templates and shared with the team: CLAUDE.md, AGENTS.md, the handoff guidelines added to every brief, and your ideal client for the client finder. Placeholders: {{project}}, {{client}}, {{country}}, {{kind}}, {{status}}, {{repository}}, {{environments}}, {{deploys}} and {{due}}.

Code and deploys

Under Settings → Connections, add your agency’s own tokens for GitHub, GitLab, Cloudflare, Vercel and Supabase. Studio uses them read-only: each project shows its pull requests, checks and deploys. Tokens stay in this Mac’s Keychain and go only to that service.

Neurova Awake and Tiles

If you use Neurova Awake or Neurova Tiles (sold separately), Studio can:

  • keep your Mac awake, lid closed too, while a connected agent works (Awake);
  • start Awake’s focus timer when you start tracking time;
  • arrange your windows with a Tiles layout after a handoff.

Studio only opens their links; it never changes their settings.

Last updated 19 November 2026

Can't find it here? See Support and FAQ or email support@neurova-ai.com. We reply by email, usually within two working days.