Try free

Last updated 13 October 2026

View a list of active MCP servers in Claude Code

Identifying connected servers in your session

You are deep into a coding task and you expect Claude Code to fetch data from your local database or your documentation server. When you ask it to use a specific tool provided by a Model Context Protocol server and it claims the tool does not exist, your workflow breaks. This often happens because the server failed to start or the configuration is not being picked up from the directory you are working in. Knowing how to quickly list every active server and its status saves you from guessing whether the issue lies in your prompt or your environment.

If you use a tool like adhdTerminal to manage multiple concurrent tasks, you can see the status of different environments across your project wall. The Model Context Protocol allows the agent to reach outside of its standard environment to interact with your specific data sources. If you have configured a server to read your tickets or query a database, that server must be active for the tools to appear in the agent command set. If the tools are missing, the agent will fallback to using only its built-in capabilities like Bash or Read, which might lead to incomplete results.

Using the internal slash command for status

Inside an active Claude Code session, the fastest way to see what is happening with your external tools is the /mcp slash command. When you type this, the agent provides a status overview of all configured servers. This list includes the names of the servers defined in your configuration files and whether they are currently connected and responsive. This is your primary diagnostic tool when you are already interacting with the agent and want to verify that a specific capability is available for the next prompt.

The output of this command is designed to be concise. It shows you the health of each bridge and confirms which servers are currently providing tools to the session. If a server you expect to see is not in this list, it means the agent did not initialize it when the session started. This usually points to a configuration error in your local project directory or a syntax error in your configuration files that prevented the agent from reading the server definitions.

Inspecting tool availability and resource counts

Sometimes the slash command shows a server name but the agent still cannot use the tools associated with it. In these cases, you should look at the specific tools the server provides. The /mcp output typically shows the count of tools or resources exposed by each server. If a server is listed but shows zero tools, there might be a runtime error in the server script or an issue with the environment variables required by that specific server. You can also use the /status command to get a broader view of the session health, which includes general information about the current environment.

If the tool count is correct but the agent refuses to use them, the issue might be related to your current permission mode. Claude Code operates in different modes that control how much freedom the agent has to run commands and tools. You can see your current mode in the terminal interface or by checking the status. If the agent is restricted, it might avoid using certain MCP tools that it perceives as high risk or outside its current instructions.

Managing servers from the command line interface

If you are not currently inside a session, you can manage and list your servers from your standard terminal shell. Running the command claude mcp opens an interactive interface specifically designed for server management. This interface allows you to view all configured servers across your system and within your current project directory. It provides a more detailed view than the internal slash command, often showing the paths to the executable files or the commands used to launch each server. This is the place to go when you need to add, remove, or repair a connection before you start a new coding session.

This external command is particularly useful for debugging the installation of a new server. It can help you verify if the server executable is reachable and if the required arguments are correctly passed. Because it runs outside of a specific agent session, it gives you a clean view of the global and local configuration without the overhead of the conversation history. Use this command whenever you make changes to your environment variables or install a new MCP server to ensure everything is ready for the next time you run the claude command.

Verifying the JSON configuration files

The configuration for these servers is usually stored in a file named .mcp.json within your project root or in your home directory. If you find that a server is missing from the list when you run claude mcp, you should verify the contents of this JSON file. Claude Code looks for this file to understand how to spawn the server processes. If the file is malformed or located in the wrong directory, the servers will not appear in the active list. You should ensure that the paths to your server scripts are absolute or correctly relative to where the .mcp.json file sits.

It is also possible to have a global configuration that applies to all your projects. This is typically found in the ~/.claude/ directory. If you have servers that you want available in every single coding session, such as a web search tool or a memory management tool, placing their configuration in the global directory is more efficient than recreating it in every repository. When listing your servers, Claude Code aggregates these different configuration sources to present a unified list of available tools.

Maintaining context across multiple projects

Managing multiple projects often means dealing with different sets of MCP servers for each repository. If you use a standard terminal, switching between these environments and keeping track of which servers are active can become a cognitive burden. This is where adhdTerminal provides a distinct advantage by letting you run many Claude Code sessions side by side as tiles on a wall. Each tile can represent a different project with its own .mcp.json configuration, and the visual feedback on each tile helps you see at a glance which sessions are working or waiting for you.

This setup prevents the confusion that arises when you have multiple hidden terminal tabs and forgot which one has access to your local database tools or specific API documentation. By having each project in its own tile, you can maintain different MCP environments simultaneously without them interfering with each other. The persistent nature of these tiles means that if you restart your computer, adhdTerminal reopens everything in the same place and resumes the same sessions, keeping your server statuses and tool availability exactly where you left them.

Troubleshooting with diagnostic utilities

When a server appears in the list but fails to function, the claude doctor command is your next step. This utility checks your entire installation, including the health of your connections and the basic requirements for running the agent. You can also run /doctor from within a session to perform these checks without quitting your current work. This can identify if the issue is a simple connection problem or something deeper like a missing dependency in your shell environment. If the doctor command suggests an update, you can use claude update to ensure you have the latest protocol handling improvements.

Sometimes the problem is not with the server itself but with how the agent perceives its environment. The claude doctor check evaluates whether the agent has the necessary permissions and path access to execute the scripts defined in your .mcp.json file. If you see warnings about path resolution or missing binaries, you must fix those at the system level before the server will show up as active and healthy. A clean report from the doctor command usually indicates that the issue lies within the specific logic of the MCP server or the prompt you are using.

Adjusting permissions for MCP tools

Keep in mind that permissions play a role in how these servers interact with your files. If you are in plan mode, the agent might be able to see the tools but will not execute them until you switch to a more permissive mode or approve the specific tool use. You can cycle through modes like default, acceptEdits, and bypassPermissions by pressing Shift+Tab. If you want to change how permissions are handled for a specific MCP tool permanently, you can modify the permissions section in your .claude/settings.json file. This ensures that your listed servers are not just visible but actually useful for the tasks you need to perform.

You can also define these permissions in .claude/settings.local.json if you want to keep your personal security preferences separate from the tracked project settings. This is useful when working in a team where everyone might have different local MCP servers or security requirements. By fine-tuning these settings, you ensure that the agent can use the tools listed in your MCP status without asking for permission on every single call, which is vital for a smooth developer experience.

Refreshing the tool list and session state

If you need to refresh the list of tools because you modified your configuration file, you do not always have to restart the entire application. While some changes might require a fresh start, you can use the /clear command to reset the current session context within the same folder. This can sometimes trigger a reload of the local environment settings and the associated tools. If that fails, quitting with Ctrl+C and restarting with claude --continue is the most reliable way to ensure that any changes to .mcp.json or your global settings are fully recognized by the agent.

For more complex changes involving multiple servers, it might be necessary to restart the background processes that manage the protocol. Claude Code handles much of this automatically when you start a new session. If you are unsure if the current list is accurate, the /status command provides a snapshot of the current state, including the active model and any relevant hooks that might be affecting tool execution. Regularly checking this status helps you stay aware of the environment the agent is operating in, ensuring that your external data sources are always ready to be queried.

Questions people ask

How can I add a new server to the list?

You can add a new server by editing the .mcp.json file in your project root or the global version in ~/.claude/. Alternatively, you can use the interactive claude mcp command from your shell to add and configure new servers through a guided interface. This allows you to define the name, executable path, and any necessary arguments for the server.

Why does a server show up in the list but its tools are not working?

This often happens if the server script has a runtime error or if the environment variables it requires are missing from your current shell. You should run claude doctor to check for general installation issues or check the terminal output for specific error messages from the server process. Verifying that the server command works independently in your terminal can also help isolate the problem.

Can I limit which tools a server provides to a specific session?

Yes, you can control tool availability by using different configuration files for different projects or by adjusting the permissions settings in .claude/settings.json. You can also switch to plan mode using Shift+Tab to review every tool call before the agent executes it. This ensures you have full control over the external data access for each specific session.

Read next

More on claude code basics·All posts