Claude Connectors in the CLI: What Carries Over

Claude connectors can carry into the CLI: Claude Code automatically makes account-linked connectors available when its active sign-in uses your claude.ai subscription. Locally configured MCP servers are a separate route, so check the current session rather than assuming every desktop connection transfers. fenbs keeps project context outside the client, letting supported assistants resume from the same work record.

How can you check your CLI connectors now?

Start in the folder where you intend to work. The goal is to identify the active account, see the actual connector list, and make one verifiable read. If you are new to the terminal client, what Claude Code is explains its role before you change any settings.

1. In your shell, run claude auth status --text. Confirm the sign-in is the account you intended to use. The CLI reference documents this status command.
2. Start Claude Code with claude. Run /status inside the session to check its active authentication method and configuration.
3. Run /mcp inside Claude Code. Find the connector you need and inspect its source and status; account-linked entries are identified as coming from claude.ai.
4. If an account connector is missing, check that it was connected in Claude’s account settings and that this session uses the appropriate account sign-in. Review organization restrictions before adding a second definition.
5. Ask for one named test record through that connector. Check the returned source against the original item before requesting any action that changes it.

Use Anthropic’s CLI reference↗ (opens in a new tab) for shell command syntax. A shell command such as claude auth status runs before or outside a session; slash commands such as /mcp run inside one. Mixing those two places is an easy reason for a correct-looking instruction to fail.

Which connectors carry over from your Claude account?

Anthropic’s MCP documentation↗ (opens in a new tab) confirms that connectors added and authenticated on claude.ai become available in Claude Code. This applies when the terminal session actively uses a claude.ai subscription login. API-key, provider, and setup-token authentication do not fetch those account connectors.

A successful model response proves you can use Claude, but it does not prove which account connection is active. Anthropic’s authentication guide↗ (opens in a new tab) explains the supported sign-in routes. Check the route used by the terminal you are actually running, especially on a machine shared between personal projects and organization work.

Suppose your browser can find a project document but the terminal cannot. First compare account identity and workspace, then connector status. Do not conclude that a second browser authorization is needed until you know whether the terminal can load that account’s connectors at all.

The general Claude connectors guide covers their purpose. For this CLI workflow, the useful evidence is the connector entry and a successful source read in this session, not its appearance in a different client.

What stays in local MCP configuration?

MCP means Model Context Protocol, a standard way for an AI app to use an outside tool. Claude Code also accepts servers configured directly in the CLI. Local scope applies privately to the current project, project scope uses a shareable .mcp.json file, and user scope applies across your projects. These definitions are distinct from account connectors.

claude mcp list
claude mcp get your-server-name

Replace your-server-name with the actual name you configured. To add a remote server, follow that service’s current Claude Code instructions, including its exact endpoint and authentication method. Do not substitute an ordinary website address or copy a credential-bearing configuration from someone else’s machine.

Sharing configuration tells teammates where a server is; it does not establish that every teammate should have the same credentials. Keep secrets out of checked-in files. Also distinguish the server’s installation scope from general settings files: Anthropic’s settings guide↗ (opens in a new tab) explains configuration locations and precedence.

If you need a connection you must build yourself, use the separate creating Claude connectors guide. Installing an existing server and designing a new one are different jobs.

Why can the same connector work in one session and fail in another?

Check the smallest difference first: account, folder, definition, or policy. A directly configured server can hide an account connector pointing at the same URL. Organization controls can restrict tools. A disabled connector or an expired authorization can also leave you with a familiar name but no usable action.

The settings guide explains that a restrictive disableClaudeAiConnectors setting can prevail over another settings source. Inspect the effective configuration instead of repeatedly editing a lower-priority file. In an organization, ask the administrator which approved route should be used; adding a duplicate is unlikely to clarify the problem.

• Missing from /mcp: check active authentication, the current project, and whether account connectors were disabled.
• Listed but unable to read: check sign-in, source account access, and the specific tool’s error.
• Reads the wrong item: check workspace, project, and source identifier instead of relying only on a repeated title.
• Works only in one folder: compare installation scope and that folder’s settings before assuming the service is unavailable.

How do you test access without granting too much?

Use a test item whose contents you already know, and request only the fields needed to confirm it. Ask Claude to identify the tool it used and return a source link or record identifier. Open the original yourself. A fluent summary without a tool result is not a connection test.

Use the connected project tool to read the task named CLI connection check.
Return its identifier, title, and current state. Do not change anything.
If you cannot reach it, report the tool error and stop.
Do not replace missing source information with a guess.

Review Claude Code’s permission controls↗ (opens in a new tab) before allowing changes. Permission rules can target particular MCP tools; broadly approving a server grants a wider set of actions. A prompt that says not to edit is a useful instruction, but it does not remove write capability from the underlying service account.

Keep the read test separate from an approved write test, and inspect both results in the source system. When continuing a shared project, have the assistant read fenbs context first so successful tool access does not become permission to reopen an already settled decision.

How to do this with fenbs

1. Create a free account and add the project you want to continue between Claude conversations and terminal sessions.
2. Connect Claude Code to fenbs through Integrations, using its browser sign-in. Check the connection in the session you will actually use.
3. Add the current tasks and record the person’s Decisions and rules. Save relevant context notes without including credentials or private source exports.
4. Ask the assistant to read project context before working, then save task progress and Where we left off before it stops. Start the next client by reading that same record.

fenbs does not copy your Claude chats, configure unrelated connectors, or make every client’s tools identical. It is not the right fit when all you need is a local MCP server with no ongoing project context to share.

What should you verify before changing clients?

Separate work continuity from tool availability. The next assistant needs the project’s approved goal and remaining checks; it also needs a working connection of its own. Our Claude Code client comparison covers interface choices. Use the connect AI topic hub for the next setup, and repeat the small read test there. fenbs Where we left off keeps the checked next step available when you switch clients.