The Best AI Workbench is Not an IDE: Building a Personal Agent Stack
These articles are AI-generated summaries. Please check the original sources for full details.
The best AI workbench is not an IDE
Engineers often mistake a single editor for an entire AI strategy rather than a window into an execution surface. A critical failure point identified in March 2026 revealed that Cursor CLI could not handle OAuth for remote MCP servers like Figma, despite native IDE support.
Why This Matters
The technical bottleneck in AI development is no longer model quality, but the agent’s ability to reach authenticated systems without functional breakage. When a client fails to handle OAuth correctly, as seen with Cursor CLI’s interaction with Figma’s remote MCP server, the integration becomes functionally unavailable, forcing a shift from all-in-one IDEs to federated workbenches that treat authentication as a primary concern.
Key Insights
- Cursor-agent OAuth failure for remote MCP servers like Figma and GitHub was reported from January 21 to March 18, 2026.
- Figma’s remote MCP server explicitly requires OAuth and does not support Personal Access Tokens (PAT), as confirmed by support in November 2025.
- Codex CLI handles remote MCP with OAuth cleanly for tools like Figma and Notion, making it a more reliable primary agent runtime than IDE-locked alternatives.
- IntelliJ IDEA remains the superior surface for large codebase navigation and refactoring, even if it is not the primary agent pilot.
- The AGENTS.md bridge allows teams to preserve Cursor-specific rules (.mdc) and skills while using external agent runtimes like Codex.
- JetBrains AI Assistant supports MCP over STDIO, Streamable HTTP, and SSE, but lacks official setup documentation from key vendors like Figma as of March 2026.
Working Examples
Adding a Figma remote MCP server to the Codex CLI runtime.
codex mcp add figma --url https://mcp.figma.com/mcpPractical Applications
- Use Case: Deploy Codex CLI as the primary agent runtime to manage Figma and Notion integrations that require stable OAuth flows while keeping IntelliJ as the core editor.
- Pitfall: Relying on Cursor IDE for source control visibility; intermittent bugs can cause agent-generated changes to fail to appear in the Git changelist.
- Use Case: Map team-wide Cursor rules located in .cursor/rules/ to non-Cursor agents to maintain institutional knowledge across different tools.
- Pitfall: Attempting to force JetBrains AI Assistant into remote MCP workflows where vendor-specific documentation and OAuth support are currently absent.
References:
- https://docs.cursor.com/cli/mcp
- https://forum.cursor.com/t/cursor-cli-figma-support/149491
- https://developers.figma.com/docs/figma-mcp-server/remote-server-installation/
- https://forum.figma.com/ask-the-community-7/support-for-pat-personal-access-token-based-auth-in-figma-remote-mcp-47465
- https://www.jetbrains.com/help/ai-assistant/mcp.html
- https://www.jetbrains.com/help/ai-assistant/use-custom-models.html
Continue reading
Next article
Top 6 Secrets Management Tools for Developers in 2026
Related Content
Beyond AI Agent Memory: The Case for Local-First Black Box Recorders
AI agent developers are shifting focus from memory to 'black box recorders' to solve critical issues like untraceable tool calls and runaway token costs.
Transform VS Code Copilot into an Autonomous AI Agent: A Technical Setup Guide
Configure VS Code Copilot as a memory-aware autonomous agent using the February 2026 v1.106 update and Model Context Protocol servers.
Building Reliable AI Agents: The 90-Day Discipline Framework
Discover why discipline, not intelligence, ensures AI agents survive the 90-day production test through mandatory state writing and identity reloading.