AI client setup

Copy-paste configuration for every client, generated from the same data as the dashboard's Connect AI page (packages/studio/client/src/connect-clients.ts). Replace /path/to/ReelForge with your ReelForge folder (on Windows write it as C:/Users/you/ReelForge). The Connect AI page (/connect) fills in the real paths, the toolsets you pick and the token when one is required.

• stdio: the client starts reelforge mcp itself. Nothing else needs to run.
• HTTP: keep reelforge studio running; the client connects to http://127.0.0.1:4310/mcp. When the dashboard binds to another address, add the header Authorization: Bearer <token> (see security).

Client	stdio	HTTP
Claude Code	✓	✓
Claude Desktop	✓
Cursor	✓	✓
VS Code · Copilot	✓	✓
Windsurf	✓	✓
Gemini CLI	✓	✓
Codex CLI	✓	✓
Zed	✓	✓
Continue	✓	✓
LM Studio	✓	✓
OpenAI-compatible		✓

Claude Code

Client documentation: https://docs.anthropic.com/en/docs/claude-code/mcp

stdio

Where: Terminal (or .mcp.json in the project, already included in this repository)

claude mcp add reelforge -- node /path/to/ReelForge/packages/cli/bin/reelforge.js mcp --root /path/to/ReelForge

Tools appear as mcpreelforge<tool>. Check with claude mcp list.

HTTP

Where: Terminal

claude mcp add --transport http reelforge http://127.0.0.1:4310/mcp

Keep reelforge studio running while you use it.

Claude Desktop

Client documentation: https://modelcontextprotocol.io/quickstart/user

stdio

Where: Settings → Developer → Edit Config (claude_desktop_config.json)

{
  "mcpServers": {
    "reelforge": {
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

Restart Claude Desktop after saving.

Claude Desktop connects to remote servers only through HTTPS connectors; use stdio locally.

Cursor

Client documentation: https://docs.cursor.com/context/model-context-protocol

stdio

Where: .cursor/mcp.json in the project (or ~/.cursor/mcp.json for every project)

{
  "mcpServers": {
    "reelforge": {
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

HTTP

Where: .cursor/mcp.json or ~/.cursor/mcp.json

{
  "mcpServers": {
    "reelforge": {
      "url": "http://127.0.0.1:4310/mcp"
    }
  }
}

VS Code · Copilot

Client documentation: https://code.visualstudio.com/docs/copilot/chat/mcp-servers

stdio

Where: .vscode/mcp.json in the project

{
  "servers": {
    "reelforge": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

Use Copilot Chat in Agent mode and enable the reelforge tools.

HTTP

Where: .vscode/mcp.json in the project

{
  "servers": {
    "reelforge": {
      "type": "http",
      "url": "http://127.0.0.1:4310/mcp"
    }
  }
}

Windsurf

Client documentation: https://docs.windsurf.com/windsurf/cascade/mcp

stdio

Where: Settings → Cascade → MCP servers → View raw config (~/.codeium/windsurf/mcp_config.json)

{
  "mcpServers": {
    "reelforge": {
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

HTTP

Where: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "reelforge": {
      "serverUrl": "http://127.0.0.1:4310/mcp"
    }
  }
}

Gemini CLI

Client documentation: https://github.com/google-gemini/gemini-cli/blob/main/docs/tools/mcp-server.md

stdio

Where: .gemini/settings.json in the project (or ~/.gemini/settings.json)

{
  "mcpServers": {
    "reelforge": {
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"],
      "timeout": 600000
    }
  }
}

Check it with /mcp inside Gemini CLI.

HTTP

Where: .gemini/settings.json or ~/.gemini/settings.json

{
  "mcpServers": {
    "reelforge": {
      "httpUrl": "http://127.0.0.1:4310/mcp",
      "timeout": 600000
    }
  }
}

Codex CLI

Client documentation: https://github.com/openai/codex/blob/main/docs/config.md

stdio

Where: ~/.codex/config.toml

[mcp_servers.reelforge]
command = "node"
args = ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
startup_timeout_sec = 30
tool_timeout_sec = 600

HTTP

Where: ~/.codex/config.toml

[mcp_servers.reelforge]
url = "http://127.0.0.1:4310/mcp"

Streamable HTTP needs a recent Codex release. With a token, export REELFORGE_TOKEN before starting Codex.

Zed

Client documentation: https://zed.dev/docs/ai/mcp

stdio

Where: Zed settings.json

{
  "context_servers": {
    "reelforge": {
      "source": "custom",
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

HTTP

Where: Zed settings.json

{
  "context_servers": {
    "reelforge": {
      "source": "custom",
      "url": "http://127.0.0.1:4310/mcp"
    }
  }
}

Continue

Client documentation: https://docs.continue.dev/customize/deep-dives/mcp

stdio

Where: .continue/mcpServers/reelforge.yaml in the project

name: ReelForge
version: 1.0.0
schema: v1
mcpServers:
  - name: reelforge
    command: 'node'
    args:
      - '/path/to/ReelForge/packages/cli/bin/reelforge.js'
      - 'mcp'
      - '--root'
      - '/path/to/ReelForge'

MCP tools are available in Agent mode.

HTTP

Where: .continue/mcpServers/reelforge.yaml in the project

name: ReelForge
version: 1.0.0
schema: v1
mcpServers:
  - name: reelforge
    type: streamable-http
    url: 'http://127.0.0.1:4310/mcp'

LM Studio

Client documentation: https://lmstudio.ai/docs/app/plugins/mcp

stdio

Where: Program → Install → Edit mcp.json

{
  "mcpServers": {
    "reelforge": {
      "command": "node",
      "args": ["/path/to/ReelForge/packages/cli/bin/reelforge.js", "mcp", "--root", "/path/to/ReelForge"]
    }
  }
}

Pick a model with tool use; a vision model can also read preview_frames images.

HTTP

Where: Program → Install → Edit mcp.json

{
  "mcpServers": {
    "reelforge": {
      "url": "http://127.0.0.1:4310/mcp"
    }
  }
}

OpenAI-compatible

Client documentation: https://spec.openapis.org/oas/v3.1.0

HTTP

Where: Any agent framework with function calling (or a GPT Action importing the OpenAPI URL)

import json, requests

API = "http://127.0.0.1:4310/api/v1"
HEADERS = {}

# Every ReelForge tool as an OpenAI function definition
tools = requests.get(f"{API}/tools?format=openai", headers=HEADERS).json()

def call_tool(name: str, arguments: str) -> str:
    r = requests.post(f"{API}/tools/{name}", json=json.loads(arguments or "{}"), headers=HEADERS)
    data = r.json()  # {"ok": bool, "result": {...}, "text": "...", "error": "..."}
    return data["text"] if data.get("ok") else "Error: " + data.get("error", "")

# OpenAPI 3.1 (one operation per tool): http://127.0.0.1:4310/api/v1/openapi.json

Pass tools to chat.completions.create(…, tools=tools) and answer each tool call with call_tool().

No MCP: uses the REST API (/api/v1) that serves the same tools.