Skip to Content
IntegrationsCursor & VS Code

Cursor & VS Code

Connect SpecForge to Cursor and VS Code with GitHub Copilot via MCP for specification-aware coding assistance.

Both Cursor and VS Code (with GitHub Copilot) support MCP servers. SpecForge integrates with both editors so you can query your specification, check ticket status, and access dependency context without leaving your development environment.

How It Differs from Claude Code

Cursor and VS Code integrations are conversational — you drive the workflow and the agent assists. You decide which ticket to work on, ask for context, implement with agent help, and mark progress.

Claude Code with Agent Teams is autonomous — you give an instruction and the agent drives the entire implementation.

Both are valid workflows. Use Cursor/VS Code when you want to stay hands-on. Use Claude Code when you want to delegate.

Setup

specforge init

The CLI detects Cursor and/or VS Code and generates .mcp.json at the repository root. Both editors read this file on startup.

Manual

Create .mcp.json at your project root:

{
  "mcpServers": {
    "specforge": {
      "command": "npx",
      "args": ["-y", "@specforge/cli@latest"],
      "env": {
        "SPECFORGE_API_KEY": "your-api-key"
      }
    }
  }
}

For Cursor: Restart the editor. You can also add via Settings > MCP in the Cursor UI.

For VS Code with Copilot: Ensure GitHub Copilot and Copilot Chat extensions are installed. Restart VS Code. SpecForge tools appear in Copilot Chat’s agent mode.

✅ The .mcp.json format is shared between Claude Code, Cursor, and VS Code. One config file works for all of them.

Available Tools

All 22 SpecForge MCP tools are available in both editors. While coding, you can:

  • Query the current ticket’s acceptance criteria and implementation context
  • Check which tickets are blocked and what to work on next
  • Query the dependency tree to understand how components relate
  • Link commits and pull requests to tickets
  • Search the specification by keyword

Workflow

A typical workflow in Cursor or VS Code with SpecForge:

  1. “What tickets are ready for implementation?” — The agent calls get_next_actionable_tickets and shows you the current wave
  2. Pick a ticket — “Give me the full context for ticket tkt_abc123” — The agent returns steps, acceptance criteria, dependencies, and blueprints
  3. Implement with assistance — Code with the agent’s help, using the ticket context as the guide
  4. Track progress — “Mark steps 1 and 2 as complete for this ticket” — The agent calls action_work_session
  5. Complete and move on — “Complete this ticket” — The agent calls complete_work_session, and dependent tickets unlock
  6. “What’s next?” — The agent shows newly unlocked tickets

The specification drives the workflow. Instead of asking “what should I build next?” to an agent with no context, you’re asking an agent that knows your entire dependency graph.

Scaffolding

To regenerate editor configuration without the full init flow:

specforge scaffold --target cursor

This creates or updates .mcp.json and generates an AGENTS.md file with instructions adapted for the editor’s agent mode.

Next Steps

  • Claude Code — For fully autonomous implementation
  • Other Agents — Gemini CLI, Codex CLI, OpenCode, and any MCP-compatible tool
Claude DesktopOther Agents