🔗

MCP Bridge

Model Context Protocol Server — 8 Workspace Tools

Glyph Editor runs a built-in MCP server that exposes your workspace as a set of structured tools. Claude Desktop, Antigravity, and Claude Code connect natively — every tool call streams into the Agent Panel's activity feed in real-time.

stdio Transport8 ToolsLive FeedWrite JournalNo Auth Required
Back to DownloadsExplorer →

// Architecture

How It Works

The MCP server is a single TypeScript file launched by each connecting client via npx tsx. No build step, no binary, no network ports — just stdin/stdout JSON-RPC.

// Connection flow
┌───────────────────┐
│ Claude Desktop │──┐
└───────────────────┘ │
┌───────────────────┐ │ ┌──────────────────────┐
│ Antigravity │──┤ │ Dragon MCP Server │
└───────────────────┘ │ │ │
┌───────────────────┐ ├──►│ 8 tools │
│ Claude Code │──┤ │ stdio transport │
└───────────────────┘ │ │ workspace-scoped │
┌───────────────────┐ │ │ │
│ Any MCP Client │──┘ └──────────────────────┘
└───────────────────┘

// Connected Engines

Native Connections

Each engine launches its own server process — they can all run simultaneously, sharing the workspace journal.

🧠

Claude Desktop

Implemented

Anthropic's native macOS application. Connects via MCP config at ~/Library/Application Support/Claude/claude_desktop_config.json.

⚡

Antigravity

Implemented

Google's agentic coding platform. Connects via MCP Servers UI or ~/.gemini/config/mcp_config.json.

💻

Claude Code

Implemented

Anthropic's CLI agent. Auto-detects the .mcp.json at the project root — zero config needed.

🔗

Any MCP Client

Open Standard

Any application that implements the Model Context Protocol can connect. No proprietary SDK required.

// Tool Reference

8 Workspace Tools

Every tool is workspace-scoped and operates through typed IPC. Read operations are safe; writes are journaled for rollback.

dragon_read_file

Read any file in the workspace — text content returned as UTF-8.

dragon_write_file

Write content to a file. Creates parent directories automatically.

dragon_list_directory

List files and subdirectories with icons and type indicators.

dragon_search_files

Grep-powered text and regex search across workspace files.

dragon_get_file_tree

Full recursive file tree with configurable depth control.

dragon_run_command

Execute shell commands in the workspace with timeout protection.

dragon_get_workspace_info

Workspace metadata — project name, path, and package info.

dragon_list_recent_writes

Journaled agent writes with prior content for safe rollback.

// Security Model

No credentials shared. No network ports. Pure stdio.
Workspace Scoped
All operations sandboxed to the project root. Paths outside are rejected.
No Credentials
Dragon Editor never holds or proxies API keys. Each engine uses its own auth.
stdio Transport
stdin/stdout JSON-RPC only. No network ports opened, no HTTP server.
Write Journal
Every agent write is journaled with prior content — human-only revert.