Forge MCP server
{
"mcpServers": {
"forgeengine": {
"type": "http",
"url": "https://mmvdabzadclebfxyzudg.supabase.co/functions/v1/forge-mcp"
}
}
}{
"mcpServers": {
"forgeengine": {
"type": "http",
"url": "https://mmvdabzadclebfxyzudg.supabase.co/functions/v1/forge-mcp"
}
}
}{
"mcpServers": {
"forgeengine": {
"type": "http",
"url": "https://mmvdabzadclebfxyzudg.supabase.co/functions/v1/forge-mcp"
}
}
}{
"servers": {
"forgeengine": {
"type": "http",
"url": "https://mmvdabzadclebfxyzudg.supabase.co/functions/v1/forge-mcp"
}
}
}{
"mcpServers": {
"forgeengine": {
"type": "http",
"url": "https://mmvdabzadclebfxyzudg.supabase.co/functions/v1/forge-mcp"
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsonversion 0.2.0
What it does
Forge Engine is an MCP connector that exposes a living design plan for teams building software with AI coding agents. Agents read specifications and milestones over MCP, claim tasks to avoid overlapping work, report changes back, and flag design drift. It is for teams coordinating multiple AI agents on shared codebases.
- Coordinate multiple coding agents working in parallel by having them claim tasks from a shared plan before starting work.
- Track design drift by flagging when code diverges from specifications or when specifications change after implementation.
- Require agents to propose changes to the design plan for human review before applying them, keeping the live plan under control.
Tools this server exposes
| Tool | Does |
|---|---|
| read_design | Retrieve the current design specification, milestones, and working context for a task. |
| claim_task | Reserve a task or system so other agents know work is in progress. |
| report_changes | Report what was built or changed back to the design plan. |
| propose_design_change | Propose a modification to the design specification for human review. |
| check_drift | Detect when code has diverged from the design or when the design has changed after implementation. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.