PixInsight MCP Bridge
Enables LLMs to interact with PixInsight's image processing capabilities through a local HTTP/SSE server, allowing listing processes, invoking them, viewing images, and more.
README
PixInsight MCP Bridge
An MCP (Model Context Protocol) bridge for PixInsight that enables LLMs to interact with PixInsight's image processing capabilities through a local HTTP/SSE server.
Overview
PixInsight MCP Bridge runs as a PixInsight script that spawns a local Node.js HTTP server implementing the MCP protocol. This allows any MCP-compatible client (Claude Desktop, VS Code Copilot, etc.) to:
- List all available PixInsight processes with categories and descriptions
- Invoke any PixInsight process with custom parameters on specific views or globally
- List open image views with metadata (dimensions, color space, bit depth)
- Get the currently focused view and its properties
- Change focus to any open view by ID
- Capture image data from any view as base64-encoded JPEG for LLM visual inspection
- Register custom processes (e.g. third-party plugins) via the dialog UI, persisted across sessions
Architecture
┌─────────────────┐ HTTP/SSE ┌──────────────────┐ stdin/stdout ┌───────────────────┐
│ MCP Client │ ◄──────────────── │ Node.js Bridge │ ◄──────────────► │ PJSR Script │
│ (Claude, etc.) │ ──────────────► │ Server │ ────────────────► │ (PixInsight) │
└─────────────────┘ └──────────────────┘ └───────────────────┘
port 3189 (default)
The bridge uses a hybrid architecture:
- PJSR Script (
pixinsight-mcp-bridge.js) — Runs inside PixInsight's JavaScript runtime (ECMA 262-5/ES5). Handles all PixInsight API interactions and spawns the HTTP server viaExternalProcess. - Node.js Bridge Server (
bridge/server.js) — Handles MCP protocol communication over HTTP. Supports both legacy SSE transport (spec 2024-11-05) and Streamable HTTP transport (spec 2025-03-26). - IPC Layer — Line-delimited JSON over stdin/stdout pipes between the two processes.
Requirements
- PixInsight 1.8.x or later
- Node.js >= 14.0.0 (must be in PATH or at a standard install location)
Installation
1. Download the plugin
Clone or download this repository:
git clone https://github.com/GaiaLabs/pixinsight-mcp-bridge.git
2. Copy to PixInsight scripts directory
Copy the src/ directory contents to your PixInsight scripts folder:
- macOS:
/Applications/PixInsight/src/scripts/MCP-Bridge/ - Windows:
C:\Program Files\PixInsight\src\scripts\MCP-Bridge\ - Linux:
/opt/PixInsight/src/scripts/MCP-Bridge/
The resulting structure should be:
PixInsight/src/scripts/MCP-Bridge/
├── pixinsight-mcp-bridge.js
├── lib/
│ └── handlers.jsh
└── bridge/
├── server.js
├── mcp-handler.js
└── ipc.js
3. Register the script in PixInsight
- Open PixInsight
- Go to Script > Feature Scripts...
- Click Add
- Navigate to the
MCP-Bridge/directory you created - Click OK — the script will appear under Script > MCP > PixInsight MCP Bridge
4. Configure your MCP client
Add the server to your MCP client configuration. For example, in Claude Desktop's claude_desktop_config.json:
{
"mcpServers": {
"pixinsight": {
"url": "http://127.0.0.1:3189/sse"
}
}
}
For clients that support Streamable HTTP transport, use:
{
"mcpServers": {
"pixinsight": {
"url": "http://127.0.0.1:3189/mcp"
}
}
}
Usage
Starting the Bridge
- Open PixInsight
- Go to Script > MCP > PixInsight MCP Bridge
- Set the port (default: 3189)
- Optionally register custom processes (see below)
- Click Start
- The bridge server will start and the MCP endpoint will be available at
http://127.0.0.1:3189
Registering Custom Processes
The built-in process registry covers ~60 common PixInsight processes. To expose third-party processes (BlurXTerminator, StarNet2, NoiseXTerminator, etc.):
- In the bridge dialog, find the Custom Processes section
- Enter the Process ID (the exact constructor name, e.g.
BlurXTerminator), a Category, and a Description - Click Add
Custom processes are persisted across PixInsight sessions. They are verified at runtime — if a registered process isn't installed, it will be silently excluded from list_processes results. You can add or remove custom processes while the bridge is running; changes take effect immediately.
Available MCP Tools
list_processes
List all available PixInsight processes and scripts.
| Parameter | Type | Required | Description |
|---|---|---|---|
category |
string | No | Filter by category (e.g. "PixelMath") |
Example response:
{
"processes": [
{ "id": "PixelMath", "category": "PixelMath", "description": "Pixel math expressions and formulas" },
{ "id": "ImageIntegration", "category": "ImageIntegration", "description": "Image stacking / integration" }
],
"count": 2
}
invoke_process
Execute a PixInsight process with specified parameters.
| Parameter | Type | Required | Description |
|---|---|---|---|
processId |
string | Yes | Process name (e.g. "PixelMath") |
parameters |
object | No | Process parameters as key-value pairs |
viewId |
string | No | View to execute on; omit for global execution |
Example — apply PixelMath to an image:
{
"processId": "PixelMath",
"parameters": {
"expression": "$T * 2",
"createNewImage": false
},
"viewId": "Image01"
}
list_views
List all open image views (main views and previews).
No parameters required. Returns view metadata including dimensions, color space, bit depth, and file path.
get_focused_view
Get the currently active/focused view and its properties.
No parameters required.
set_focused_view
Change focus to a specific view by ID.
| Parameter | Type | Required | Description |
|---|---|---|---|
viewId |
string | Yes | View ID (e.g. "Image01") |
get_image_from_view
Get the actual image contents of a view as a base64-encoded JPEG. The image is saved to a temporary JPEG file, read back as base64, and returned as MCP image content alongside metadata (dimensions, color space, bit depth).
| Parameter | Type | Required | Description |
|---|---|---|---|
viewId |
string | No | View ID to capture. If omitted, captures the currently focused view |
The response includes both the image data (as MCP image content) and a text block with view metadata.
Standalone Mode (Testing)
You can run the bridge server without PixInsight for testing and development:
node src/bridge/server.js --standalone --port 3189
This starts the server with mock responses, useful for testing MCP client integrations.
MCP Transport Support
The bridge supports two MCP transports:
Legacy HTTP+SSE (2024-11-05 spec)
GET /sse— Opens SSE stream, receivesendpointevent with message URIPOST /messages?sessionId=<id>— Send JSON-RPC messages; responses arrive via SSE
Streamable HTTP (2025-03-26 spec)
POST /mcp— Send JSON-RPC messages; response returned as JSON body or SSE streamGET /mcp— Open SSE stream for server-initiated messagesDELETE /mcp— Terminate session
Health Check
GET /health— Returns{ "status": "ok" }
Development
Project Structure
pixinsight-mcp-bridge/
├── src/
│ ├── pixinsight-mcp-bridge.js # Main PJSR entry point (ES5)
│ ├── lib/
│ │ └── handlers.jsh # PixInsight command handlers (ES5)
│ └── bridge/
│ ├── server.js # Node.js HTTP/SSE MCP server
│ ├── mcp-handler.js # MCP protocol logic
│ └── ipc.js # IPC communication layer
├── test/
│ ├── run-tests.js # Test runner
│ ├── test-mcp-handler.js # MCP protocol tests
│ ├── test-server.js # HTTP server integration tests
│ └── test-handlers.js # PJSR handler tests (mocked PI API)
├── package.json
└── README.md
Key Design Decisions
- Hybrid architecture: PixInsight's PJSR runtime has no HTTP server capability (
NetworkTransferis client-only). The solution usesExternalProcessto spawn a Node.js HTTP server and communicates via stdin/stdout IPC. - ES5 compliance: All PJSR code (
*.jsand*.jshinsrc/) is ECMA 262-5 compliant — nolet/const, no arrow functions, no template literals, no classes, no promises. - Zero dependencies: The Node.js bridge server uses only built-in modules (
http,url,crypto). No npm install required. - Process registry: Since PJSR doesn't expose a direct "list all processes" API, the bridge maintains a curated registry of known processes and verifies availability at runtime via constructor detection.
Running Tests
npm test
# or
node test/run-tests.js
The test suite includes:
- MCP protocol tests — JSON-RPC message handling, initialization, tool listing/calling
- HTTP server tests — Both SSE and Streamable HTTP transports, CORS, error handling
- Handler tests — PJSR command handlers with mocked PixInsight API
Adding New Tools
- Define the tool schema in
src/bridge/mcp-handler.jsin theTOOLSarray - Add parameter validation in
MCPHandler.prototype._validateToolParams - Implement the handler in
src/lib/handlers.jshinCommandDispatcher - Add mock handling in
createMockHandlerinsrc/bridge/server.js - Write tests
Adding New Processes to the Registry
For end users: Use the Custom Processes section in the bridge dialog UI to register additional processes. These are persisted via PixInsight's Settings API and survive restarts.
For developers: Edit CommandDispatcher.prototype._getProcessRegistry in src/lib/handlers.jsh to add new built-in process entries:
{ id: "NewProcess", category: "CategoryName", description: "What it does" }
All processes (both built-in and custom) are verified at runtime — entries for processes not installed in the user's PixInsight will be automatically excluded.
PJSR Development Notes
- Use
#include "path/file.jsh"for includes (preprocessor directive, not standard JS) - Use
#feature-id Category > Nameto register scripts in PixInsight's Script menu - All PixInsight processes are global constructors (e.g.,
new PixelMath()) - Use
processEvents()in loops to keep the GUI responsive ExternalProcessrequiresprocessEvents()polling for event handlingTimeruses seconds (not milliseconds) for its interval
Troubleshooting
"Node.js not found"
Ensure Node.js >= 14 is installed and accessible. The script searches these locations:
- macOS/Linux:
which node,/usr/local/bin/node,/usr/bin/node,/opt/homebrew/bin/node - Windows:
where node,C:\Program Files\nodejs\node.exe
Server won't start
Check the PixInsight console (View > Process Console) for error messages. Common issues:
- Port already in use — change the port number
- File permissions — ensure the bridge server files are readable
MCP client can't connect
- Verify the server is running (check
http://127.0.0.1:3189/health) - Ensure your MCP client config points to the correct endpoint
- The server only listens on
127.0.0.1(localhost) — it's not accessible from other machines
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。