The short map
| Tool | Where you add it | Who can add it | Remote or local |
|---|---|---|---|
| Claude (Pro, Max) | Customize > Connectors > Add > Add custom connector | You | Remote, HTTPS |
| Claude (Team, Enterprise) | Organization settings > Connectors > Add > Custom > Web | Owners, Primary Owners (Enterprise also custom roles with Manage on Libraries) | Remote, HTTPS |
| Claude Desktop | Remote connector through your Claude account, or a local server through desktop extensions (MCPB) or the config file | Depends on account and organization | Remote and local |
| ChatGPT | Plugins > + > Add custom MCP server | Business: admins and owners; Enterprise/Edu also members with the right role; Pro read-only | Remote |
| GitHub Copilot in VS Code | Chat: Open Customizations > MCP Servers > Add Server | You, if your org allows MCP | Remote (HTTP) or local (Command) |
| Grok Business and Enterprise | console.x.ai > Grok Business > Connectors > + Add Connector > Other | Admin with Team Read-Write | Remote |
| xAI API | server_url and server_label in the request |
Developers | Remote |
Claude
Custom connectors work on Free (one connector), Pro, Max, Ultra, Team and Enterprise, across Claude, Cowork and Claude Desktop. On Team and Enterprise an admin adds the connector once, and each member then clicks Connect under Customize > Connectors to sign in as themselves. The server must be reachable from the public internet over HTTPS on port 443; a server only on your VPN does not work unless you allowlist Anthropic’s IP ranges. Claude’s docs recommend Streamable HTTP; the older HTTP+SSE transport is being phased out. Tool results are capped around 150,000 characters and each call has a 240 second timeout.
ChatGPT
OpenAI’s docs describe adding a custom MCP server under Plugins, then the plus button, then Add custom MCP server, with the URL and authentication (OAuth). Full support is in ChatGPT Business and Enterprise/Edu on the web. On Business, admins and owners create and publish the connection; Enterprise/Edu can also authorize selected members through role-based access. Pro supports custom servers with read and fetch permissions. When ChatGPT asks for confirmation depends on the app’s permissions and the action. The search and fetch tools are no longer required; OpenAI recommends them for company knowledge. OpenAI states it does not build or review custom MCP servers.
GitHub Copilot
In VS Code, open Chat: Open Customizations > MCP Servers > Add Server and choose HTTP or Command (stdio), or run MCP: Add Server from the command palette. A .mcp.json in the repository shares the setup through git. On Copilot Business and Enterprise, the MCP servers in Copilot policy is off by default; your admin turns it on per organization. Limit tools per server with the tools field. Note that Copilot cloud agent and code review use configured MCP tools without asking for approval.
Grok
In Grok Business and Enterprise an admin adds a custom MCP server: sign in at console.x.ai, pick your team, go to Grok Business > Connectors, click + Add Connector, choose Other and enter the server URL. Members then connect their own account. A server on your own machine needs a separate tunnel.
Developers use remote MCP through the xAI API. The Responses API takes server_url, server_label and optionally headers and allowed_tools; the native xAI SDK uses extra_headers and allowed_tool_names. Without such a limit, every tool on the server enters the context. require_approval is not supported, so limit tools up front.
The misunderstanding
“Connect the server once and all our AI tools share it.” Each tool has its own place to add servers, its own admin switch, and its own sign-in. Four tools means four setups, and four places to keep in step. And what each connection sees depends on the account, storage and rights the server uses.
What this means for a team
Plan who owns each connection: the Claude owner, the ChatGPT workspace admin, the GitHub org admin, the Grok Business admin. Use per-person sign-in, not one shared key: Claude warns that a shared key on Team or Enterprise gives everyone the same rights. More on controls in is MCP safe to use in a company, and on memory servers in what an MCP memory server is.
Where Bombos fits
Bombos is a plugin that connects to each of these tools, plus Claude Code, Codex, Cursor and your own agents, and gives all of them one shared memory: team knowledge, rules the AI learned about how you work, and what already sits in mail, documents and repos, with a trail of what each answer rests on. Cloud or your own server, and one person can start with their own agents.
Related questions
Can I connect a local MCP server to claude.ai?
Not through a custom connector. Claude connects to remote servers from Anthropic's own IP ranges, so the server must be reachable from the public internet over HTTPS. For local servers, Claude Desktop uses desktop extensions (MCPB) or its config file.
Why don't I see MCP in GitHub Copilot at work?
On Copilot Business and Enterprise, the policy MCP servers in Copilot is off by default. An organization or enterprise admin has to turn it on. The policy does not apply to Copilot Free, Pro, Pro+ and Max.
Which ChatGPT plans can add a custom MCP server?
According to OpenAI's Help Center, ChatGPT Business and Enterprise/Edu have full support; on Business, admins and owners create and publish connections, and Enterprise/Edu can also let selected members do it through roles. Pro supports custom servers with read and fetch permissions only.
Can Grok use MCP servers?
Yes. In Grok Business and Enterprise an admin adds a custom MCP server in console.x.ai under Grok Business > Connectors, and members then connect their own account. Developers can also use remote MCP through the xAI API.
Sources
- Get started with custom connectors using remote MCP (Claude Help Center)
- Building connectors (Claude docs)
- Connecting custom MCP servers in ChatGPT (OpenAI Help Center)
- Use MCP servers in VS Code
- About Model Context Protocol in GitHub Copilot
- Remote MCP tools (xAI docs)
- Connector management (Grok Business docs)
