Connect other AI agents

Tested with Claude Code and Codex. The clients below use the same MCP standard; these steps follow each client's own documentation, and Shotomatic has not been tested with them. Client menus and file formats change, so check the client's docs if a step no longer matches.

Every client needs the same three values:

• Command: npx
• Arguments: -y, shotomatic, mcp
• Environment variables: none. There is no API key.

Before you add it, turn on Let AI agents use Shotomatic in Settings > AI agents, and install Node 18.3 or later. The first time a client connects, Shotomatic shows an approval card; select Allow. See Connect an AI agent for the full flow, including Claude Code, Claude Desktop and Codex.

On Windows, ~ in the paths below means your user folder (%USERPROFILE%). If a client on Windows cannot start npx, use cmd as the command and /c, npx, -y, shotomatic, mcp as the arguments.

Each client has its own MCP settings, and many ask before every tool call. Add Shotomatic where the client expects a local (stdio) server, not a remote URL.

Amp

Docs: Amp MCP. Run:

amp mcp add shotomatic -- npx -y shotomatic mcp

Run amp mcp doctor to check the server. Amp's manual says to bundle MCP servers in skills when you do not need them in every thread.

Cline

Docs: Configuring MCP servers. In the Cline panel, select the MCP Servers icon, open Configure, and select Configure MCP Servers. Add this to the JSON file that opens (the Cline CLI uses ~/.cline/data/settings/cline_mcp_settings.json):

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Cline's autoApprove list is empty by default, so it asks before each tool call.

Continue

Docs: Set up MCP in Continue. MCP works in Agent mode only. Create .continue/mcpServers/shotomatic.yaml in your workspace:

name: Shotomatic
version: 0.0.1
schema: v1
mcpServers:
  - name: Shotomatic
    type: stdio
    command: npx
    args:
      - "-y"
      - "shotomatic"
      - "mcp"

GitHub Copilot CLI

Docs: Adding MCP servers to Copilot CLI. Run:

copilot mcp add shotomatic -- npx -y shotomatic mcp

This writes to ~/.copilot/mcp-config.json. In a session, enter /mcp to open the MCP server dashboard.

Cursor

Docs: Model Context Protocol in Cursor. Add this to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json inside one project:

{
  "mcpServers": {
    "shotomatic": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Check that the server is enabled in Customize. Cursor shows start-up errors in Output (Cmd + Shift + U, or Ctrl + Shift + U on Windows) under MCP Logs. Cursor asks for approval before using MCP tools by default.

Devin (Windsurf)

Docs: Devin CLI MCP configuration and Devin Desktop customizations. Windsurf's own docs address now redirects to the Devin Desktop docs. Run:

devin mcp add -s user shotomatic -- npx -y shotomatic mcp

Or add the same JSON as above (without type) to ~/.config/devin/mcp_config.json. Run devin mcp list, or in Devin Desktop open Customize > MCPs and look under This computer. The agent asks for approval before calling any MCP tool by default.

Gemini CLI

Docs: MCP servers with Gemini CLI. Add this to ~/.gemini/settings.json (or .gemini/settings.json in a project):

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Run gemini mcp list in a shell, or /mcp inside a session. Stdio servers show as connected only when the current folder is trusted. Leave trust at its default (false) so Gemini CLI keeps asking before each tool call.

Goose

Docs: Using extensions. In Goose Desktop, open Extensions, select Add custom extension, and enter:

• Type: Standard IO
• ID: shotomatic
• Name: Shotomatic
• Command: npx -y shotomatic mcp

In the CLI, run goose configure, choose Add Extension and Command-line Extension, and enter the same command. For a single session, run goose session --with-extension "npx -y shotomatic mcp".

JetBrains AI Assistant and Junie

Docs: AI Assistant MCP and Junie CLI MCP configuration.

AI Assistant: open Settings | Tools | AI Assistant | Model Context Protocol (MCP), select Add, paste the JSON below, and select Apply. The Status column shows the connection.

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Junie: add the same JSON to ~/.junie/mcp/mcp.json (or .junie/mcp/mcp.json in a project), then run /mcp to confirm the server is active.

Kiro

Docs: Kiro MCP configuration. Run Kiro: Open user MCP config (JSON) from the Command Palette, or edit ~/.kiro/settings/mcp.json (.kiro/settings/mcp.json for one workspace). Make sure the MCP support setting is on, then add:

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Saving the file reconnects the servers.

LM Studio

Docs: Use MCP servers in LM Studio. LM Studio supports local MCP servers and follows Cursor's mcp.json notation. Open the Program tab in the right sidebar and select Install > Edit mcp.json. Add:

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

LM Studio's docs advise against installing MCP servers from sources you do not trust and warn that some servers use many tokens, which can overflow a local model's context.

OpenCode

Docs: OpenCode MCP servers. Run this in the project that should use it, and add --global to make it available everywhere:

opencode mcp add shotomatic -- npx -y shotomatic mcp
opencode mcp list

To edit the config by hand in the current (V2) docs, add this to opencode.jsonc. The V1 format is different, so use the CLI command if you are not sure which version you have.

{
  "mcp": {
    "servers": {
      "shotomatic": {
        "type": "local",
        "command": ["npx", "-y", "shotomatic", "mcp"]
      }
    }
  }
}

Qwen Code

Docs: Connect Qwen Code to tools via MCP. Add this to ~/.qwen/settings.json (user) or .qwen/settings.json (project):

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

Restart Qwen Code in the same project, then enter /mcp to see the server.

Roo Code

Docs: Using MCP in Roo Code. Open the MCP settings view from the icon at the top of the Roo Code pane, scroll down, and select Edit Global MCP (mcp_settings.json) or Edit Project MCP (.roo/mcp.json). Add:

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

VS Code with GitHub Copilot

Docs: MCP configuration reference and Manage MCP servers. VS Code uses servers, not mcpServers, and requires type. Add this to .vscode/mcp.json in a workspace:

{
  "servers": {
    "shotomatic": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

To add it to your user profile instead, run:

code --add-mcp '{"name":"shotomatic","type":"stdio","command":"npx","args":["-y","shotomatic","mcp"]}'

Run MCP: List Servers from the Command Palette to see the server and start or restart it. VS Code may ask you to trust the server the first time it starts.

Warp

Docs: Model Context Protocol in Warp. Open Settings > Agents > MCP servers, select + Add, and paste:

{
  "mcpServers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"]
    }
  }
}

The MCP servers page lists each server and whether it is running. Warp's agent asks before calling MCP servers unless you allow a server in the agent profile.

Zed

Docs: Model Context Protocol in Zed. Zed calls them context servers and uses context_servers. Run zed: open settings file from the Command Palette and add:

{
  "context_servers": {
    "shotomatic": {
      "command": "npx",
      "args": ["-y", "shotomatic", "mcp"],
      "env": {}
    }
  }
}

Open Settings > AI > MCP Servers: a green dot with the tooltip "Server is active" means it is connected. Zed asks before tool calls by default.

Limits to know about

• Shotomatic returns screenshots as images inside tool results. Cursor and Gemini CLI document support for images in MCP tool results. For the other clients above, the docs do not say, so the agent may receive only text.
• Several clients ask for approval on every tool call by default, which can mean many prompts during a long capture. Each client's docs describe how to allow a server's tools; see Limits and safety for what Shotomatic itself allows agents to do.

Troubleshooting

• The client does not see the server. Save the config file, then restart the client or reload its MCP list. Check that the entry sits under the key that client expects (mcpServers, servers or context_servers) and that the JSON or YAML is valid.
• "MCP access is turned off." Open Shotomatic, go to Settings > AI agents, turn on Let AI agents use Shotomatic, and restart the client.
• No approval card appears. Open Shotomatic and look for the card; on a Mac, its Dock icon bounces when the card appears. Until you approve, the agent can only read status. Removing the agent in Settings > AI agents brings the card back on the next connection.
• The client cannot find npx or Node. On a Mac, apps you open from the Dock or Finder often start with a shorter PATH than Terminal. Cursor's docs say the command must be on the system path or given as a full path, and Qwen Code's docs recommend an absolute command path when a server fails to start. In Terminal, run which npx and use that full path as the command, for example /opt/homebrew/bin/npx. npx also needs node on its PATH; if it still fails, add the folder from which node to a PATH entry under the server's env (where the client supports env). Version managers such as mise or asdf may need the same treatment.

Back to Connect an AI agent.