Skip to content

Connect a coding agent

Coding agents work with Neova through its MCP server. Each Neova window runs its own server for the project it has open. An agent reaches that server through a small connector script that ships with Neova. The connector is listed as an MCP server in the agent’s configuration file.

Set it up in Settings › Server, or in the Welcome to Neova window the first time you open a project.

The server card at the top of Settings › Server shows the state of the server.

State What you see
Server running The port and the number of connected agents, for example Port 21700 · 2 connected agents.
Server stopped A Port (leave empty for auto): field. Type a port, or leave it empty to let Neova pick one.
Button What it does
Projects Opens the list of known projects and the ports they use.
Start Server / Stop Server Starts or stops the server for this window.

Neova writes the port the server runs on to .neova/mcp-port in the project folder. The connector reads this file each time it connects, so the agent finds the server even when the port changes.

The agent’s configuration needs one entry, named neova, under mcpServers:

{
"mcpServers": {
"neova": {
"command": "node",
"args": [
"/Applications/Neova Studio.app/Contents/Resources/mcp-connector/index.js",
"--project",
"/Users/you/code/my-project"
]
}
}
}
Field Value
command node
args[0] The path to Neova’s connector script. Copy it from the status bar: hover over MCP and click Copy connector path.
--project The project folder. The connector connects to the Neova window that has this folder open.
env.NEOVA_MCP_KEY Only present when Require access keys is on in Settings › Server. Neova then writes the key into this entry.

When Neova writes the entry, it merges it into the file. Other servers in the file stay as they are. If the file is not valid JSON, Neova does not change it and shows an error.

If the project is not open in Neova, the agent’s connector logs <folder> is not open in Neova. Open that project and restart this agent.

The Welcome to Neova window shows the first time you use Neova.

  1. Under 1. How do you start your coding agent?, type the command you use to start your agent, for example claude.
  2. Under 2. Let it reach Neova, click Add Neova to .mcp.json. When the entry is in place, the window shows Already set up.
  3. Click Done.

If the server is not running, the window says so, and you can add the entry later from Settings › Server.

  1. Go to Settings › Server.
  2. Under Agent Configuration, choose the tab for your agent.
  3. For Claude Code and Cursor, click the button in the card below the configuration. For the other tabs, click Copy and paste the configuration into the agent’s file yourself.
Tab Target file Add from Neova
Claude Code .mcp.json Yes
Cursor .cursor/mcp.json Yes
Gemini .gemini/settings.json No, copy the configuration
Mistral Vibe .vibe/mcp.json No, copy the configuration
OpenCode .opencode/mcp.json No, copy the configuration

The card shows the state of the target file in the project folder:

Card title Button
Configure the MCP connection in this project directory Add to Project. The file does not exist yet; Neova creates it.
Add Neova to your existing MCP configuration Add Neova. The file exists and the note (file exists, will merge) shows.
Update Neova in your MCP configuration Update. The entry exists but points to another connector path or project folder.
Neova is configured in this project Configured. Nothing to do.

On the Claude Code tab, a second card, Report agent status for Claude sessions you start yourself, adds status hooks to .claude/settings.local.json. Click Add to settings. Agents that Neova starts report their status without it.

Restart the agent after you add or update the entry, so it reads the new configuration. Claude Code asks whether to use the servers in .mcp.json; approve neova.

An agent that you start from Neova, on Agents › Team or Agents › Sessions, runs in a terminal that Neova opens. Neova sets two environment variables in that terminal:

Variable Holds
NEOVA_AGENT_ID The ID of the agent the terminal belongs to.
NEOVA_TERMINAL_TOKEN A secret for that one terminal.

The agent passes both on to the connector. Neova checks the token and links the connection to that agent. The agent then works under the name it has on Agents › Team.

An agent that you start in a terminal outside Neova connects as an external agent. Its name gets the suffix (external), for example claude-code (external). It can use the tools, but it is not linked to an agent on Agents › Team.

An agent can call register_agent with a preferredName to choose the name it shows in Neova. Without a call, Neova gives it a generated name.

The result contains success and name. When the name is already in use by another connected agent, success is false and the agent keeps its current name. An agent that is linked to an agent on Agents › Team keeps that agent’s name; rename it in the Agent Editor instead.

Where What to look for
Status bar, MCP 2 agents · :21700 when the server runs, off when it is stopped. Hover for the port and the names of the connected agents, plus Copy connector path and Restart. Click it to open Settings › Server.
Agents › Team A green link icon next to the agent’s name. Hover over it to see Linked to and the name of the connection.
Agents › Sessions A green link badge on the agent’s avatar, with the same Linked to tooltip.
Settings › Server The number of connected agents in the server card.

If an agent does not show up:

  1. Check that the MCP item in the status bar is not off. If it is, click Restart in its hover card, or Start Server in Settings › Server.
  2. Check that the card under Agent Configuration says Neova is configured in this project. If it says Update Neova in your MCP configuration, click Update.
  3. Restart the agent.