Claude Desktop
Status: confirmed
Desktop AI workflows connected to approved local tools and web/data connectors.
Remote connector and enterprise policy behavior can differ from local desktop setup.
Official sourceMCP in AI clients
Compare how MCP usage differs across AI clients, editors, and agent frameworks. Client support and setup details change quickly, so every row includes a current official source.
Short answer
Claude Desktop, Claude Code, Cursor, VS Code Copilot, and Windsurf/Devin Desktop have documented MCP support. OpenAI support is strongest for developer-built agents and apps, not generic consumer local MCP installation.
| Client/environment | Support status | Setup method | Global vs project | Config location or UI | Local vs remote | Auth handling | Secret handling | Tool approval UX | Best use case | Limitations | Source |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Claude Desktop | confirmed | Local MCP server configuration documented by the MCP project. | Usually app-level local configuration. | Client configuration file; verify the current platform path before editing. | Local stdio servers are the common documented pattern. | Handled by each server through environment variables, OAuth, or local credentials. | Keep secrets outside committed project files; use local environment values where supported. | Client-level tool discovery and approval behavior can change; verify in the current Claude UI. | Desktop AI workflows connected to approved local tools and web/data connectors. | Remote connector and enterprise policy behavior can differ from local desktop setup. | MCP user quickstart |
| Claude Code | confirmed | Claude Code documents connecting MCP servers from the command line and project context. | Local, project, and user-level scopes are documented in Claude Code materials. | Claude Code command/config flow; check current docs before writing config by hand. | Supports local and remote server patterns documented by Claude Code. | Depends on server transport and auth method. | Use local secrets or environment variables; avoid committing keys. | Claude Code prompts around tool use and permissions inside coding workflows. | Technical operators building competitor-tracking workflows inside repos or CLI agents. | Best suited to technical users. | Claude Code MCP documentation |
| Cursor | confirmed | Cursor's official docs list MCP servers as part of Cursor customization. | Workspace and user-level behavior should be checked in current Cursor docs. | Cursor docs route MCP information through the main docs/customization area. | MCP server support is documented, but exact transport details should be verified per server. | Server-specific; API keys and OAuth are handled by MCP server configuration. | Keep secrets out of shared repository files and check Cursor's current config guidance. | Cursor exposes tools in the coding agent experience; approval behavior may vary by version. | Developer and technical marketer workflows that combine competitor data with code, docs, or local files. | Cursor docs structure changes; use official docs before copying a config path. | Cursor official documentation |
| VS Code Copilot | confirmed | VS Code documents MCP server setup for agent customization. | User and workspace/project configuration are documented by VS Code. | VS Code agent customization MCP server docs. | Local server and configuration patterns are documented; confirm transport support per server. | Server-specific, with secrets handled through VS Code and the operating environment. | Use VS Code secret guidance where available; do not commit API keys. | Copilot agent/tool approval UX applies inside VS Code. | Teams already using VS Code and GitHub Copilot for technical workflows. | Availability can depend on Copilot plan, VS Code version, and organization settings. | VS Code MCP servers documentation |
| Windsurf / Devin Desktop | confirmed | The Windsurf MCP page currently redirects to Devin Desktop MCP documentation. | Desktop/client configuration; verify current app behavior and organization policy. | Devin Desktop MCP documentation. | Client MCP support is documented; server transport details should be checked per connector. | Server-specific; API keys and OAuth are set through the documented configuration flow. | Use approved secret handling and avoid shared plain-text keys. | Agent approval UX depends on the current Devin Desktop/Windsurf build. | Agentic coding or operations workflows from a desktop AI environment. | Branding and docs location changed; old Windsurf URLs can redirect. | Devin Desktop MCP documentation |
| OpenAI Responses API | developer-framework | OpenAI documents MCP tools/connectors for developer-built applications. | Application-level configuration, not a consumer desktop mcp.json file. | Code/API request configuration. | Remote MCP connector pattern is documented for OpenAI platform use. | Developer application controls auth and connector access. | Use application secret management and keep connector credentials server-side. | Developer-defined application UX plus OpenAI platform tool behavior. | Custom competitive-intelligence agents and internal apps. | Not the same as adding arbitrary local MCP servers to the ChatGPT consumer UI. | OpenAI MCP tools and connectors guide |
| OpenAI Agents SDK | developer-framework | Agents SDK documentation includes MCP integration for custom agents. | Application or agent-level code configuration. | Agent codebase. | SDK-level MCP support; exact server transport depends on implementation. | Controlled by the agent application and MCP server. | Use environment variables or managed secrets in the agent runtime. | Developer-defined. | Custom CLI agents, background monitors, and scheduled competitor workflows. | Requires engineering work; not a plug-and-play marketer tool. | OpenAI Agents SDK MCP docs |
| ChatGPT custom apps | partial | OpenAI Apps SDK uses an MCP server for ChatGPT app backends. | App/server configuration, not a general local MCP client setting for every ChatGPT user. | Apps SDK server code and app manifest flow. | MCP server concepts are part of the Apps SDK flow. | App-specific OAuth/auth flows where implemented. | Handled by the app backend and deployment environment. | ChatGPT app UX, not generic local tool approval. | Building a packaged ChatGPT app around a specific competitive-intelligence workflow. | Do not assume ChatGPT consumer clients can install every local MCP server directly. | OpenAI Apps SDK MCP server concept |
Client-by-client notes
Claude Code
Status: confirmed
Technical operators building competitor-tracking workflows inside repos or CLI agents.
Best suited to technical users.
Official sourceCursor
Status: confirmed
Developer and technical marketer workflows that combine competitor data with code, docs, or local files.
Cursor docs structure changes; use official docs before copying a config path.
Official sourceVS Code Copilot
Status: confirmed
Teams already using VS Code and GitHub Copilot for technical workflows.
Availability can depend on Copilot plan, VS Code version, and organization settings.
Official sourceWindsurf / Devin Desktop
Status: confirmed
Agentic coding or operations workflows from a desktop AI environment.
Branding and docs location changed; old Windsurf URLs can redirect.
Official sourceOpenAI Responses API
Status: developer-framework
Custom competitive-intelligence agents and internal apps.
Not the same as adding arbitrary local MCP servers to the ChatGPT consumer UI.
Official sourceOpenAI Agents SDK
Status: developer-framework
Custom CLI agents, background monitors, and scheduled competitor workflows.
Requires engineering work; not a plug-and-play marketer tool.
Official sourceChatGPT custom apps
Status: partial
Building a packaged ChatGPT app around a specific competitive-intelligence workflow.
Do not assume ChatGPT consumer clients can install every local MCP server directly.
Official sourceBest option by use case
Desktop operator
Start with Claude Desktop for low-risk local tools, then keep sensitive browser or paid API servers project-scoped.
Developer workflow
Use Claude Code, Cursor, VS Code, or Windsurf/Devin Desktop depending on the editor your team can secure.
Custom agent
Use OpenAI Responses API, OpenAI Agents SDK, or another agent framework when you need an internal productized workflow.
Head-to-head comparisons
MCP client comparison
Claude Desktop vs Cursor for MCP
Compare Claude Desktop and Cursor for MCP setup, scope, secret handling, tool approval, and competitor-tracking use cases.
MCP client comparisonCursor vs VS Code for MCP
Compare Cursor and VS Code Copilot for MCP setup, project scope, developer workflows, and competitive-intelligence tasks.
MCP client comparisonChatGPT vs Claude for MCP
Compare ChatGPT-related MCP support with Claude Desktop and Claude Code for setup and competitive-intelligence workflows.
MCP client comparisonWindsurf vs Cursor for MCP
Compare Windsurf/Devin Desktop and Cursor for MCP workflows, setup caveats, and technical competitive-intelligence work.
AI-client source citations
Use these links to verify setup, pricing, support, and current product behavior before installing anything.
- MCP user quickstart
Documents user-facing MCP setup concepts and local client configuration patterns. Last checked 2026-06-29.
- Claude Code MCP documentation
Official Claude Code MCP documentation. Last checked 2026-06-29.
- Cursor official documentation
Official Cursor documentation covering Agent, Rules, MCP, Skills, and CLI. Last checked 2026-06-29.
- VS Code MCP servers documentation
Official VS Code documentation for MCP server configuration. Last checked 2026-06-29.
- Devin Desktop MCP documentation
Official redirected documentation for Windsurf/Devin Desktop MCP setup. Last checked 2026-06-29.
- OpenAI MCP tools and connectors guide
Official OpenAI documentation for MCP tools and connectors in developer workflows. Last checked 2026-06-29.
- OpenAI Agents SDK MCP docs
Official Agents SDK documentation for connecting agents to MCP servers. Last checked 2026-06-29.
- OpenAI Apps SDK MCP server concept
Official Apps SDK documentation describing MCP server concepts for ChatGPT apps. Last checked 2026-06-29.
MCP client comparison FAQ
Which AI clients have confirmed MCP support?
This page only includes environments with current official documentation or official developer docs mentioning MCP support or MCP-server concepts.
Does ChatGPT support local MCP servers like Claude Desktop?
Do not assume that. OpenAI documents MCP through platform tools, Agents SDK, and Apps SDK concepts. That is different from a general local desktop MCP client.
Should I use global or project-level MCP config?
Use project-level config for sensitive servers, paid APIs, browsers, mailboxes, local files, or workflow-specific data. Global config is best for low-risk utilities.