GitHub Copilot Agent Mode Not Showing or Not Working: Fixes

When GitHub Copilot agent mode is missing or not working in VS Code, work through it in this order: update VS Code, check you are signed in with the GitHub account that holds your Copilot plan, check the chat.agent.enabled setting and whether your organisation manages it, then look at which session target you picked, because Agent is now a role inside a Local session rather than a mode of its own. If Agent is there but runs stall, look at your AI credits, your extensions and your MCP servers. Each step is below, with the command that tells you whether it is the cause.

Match the symptom to the section

• No Agent in the picker at all: the setting, a policy, or a session target that does not offer it.
• Chat itself is missing or greyed out: the account, an old VS Code, or AI features turned off.
• Agent starts, then stops with a message about limits: AI credits or a temporary rate limit.
• Agent works, but a tool or MCP server is not there: the session target or an MCP policy.
• Everything looked fine yesterday: an extension update, or a VS Code update you have not restarted into.
• You are not in VS Code: JetBrains, Visual Studio and Xcode put agent mode in a different place.

1. Update VS Code and check the account

VS Code’s troubleshooting guide↗ (opens in a new tab) starts with the same basic checks every time. Run Code: Check for Updates from the Command Palette and restart. Then confirm VS Code is signed in with the GitHub account that has your Copilot plan; a personal account signed in where your work account should be is a common reason agent mode appears on one machine and not another. To switch, sign out from the Accounts menu in the Activity Bar and sign back in.

While you are there, open the Copilot status dashboard from the Copilot icon in the Status Bar. It shows whether you have reached an allowance. And reproduce the problem in a new chat session before changing anything else, keeping the prompt and the model the same, so you know it is not one long session that has gone wrong.

2. Agent is missing: the setting and the policy

The switch for agents still exists. VS Code’s chat FAQ↗ (opens in a new tab) says to check that agents are turned on with chat.agent.enabled, that your organisation can manage it with a policy, and that if the setting is unavailable or managed you should contact your administrator. Check your workspace settings as well as your user settings, because a committed .vscode/settings.json can turn it off for everyone who opens the repository.

{
  "chat.agent.enabled": true,
  "chat.disableAIFeatures": false
}

chat.disableAIFeatures is the other one to look for. When it is on, VS Code hides chat and inline suggestions and disables the Copilot extensions, so the whole panel disappears rather than just Agent. If a setting shows as managed and you cannot change it, run Developer: Policy Diagnostics. It shows the effective value of each policy and where it came from. According to VS Code’s enterprise AI settings page↗ (opens in a new tab), the policy behind chat.agent.enabled is called ChatAgentMode, which is the name to give your IT team.

3. Where agent mode went: session targets

Older guides tell you to pick Agent from a dropdown of modes. VS Code now starts every chat session with a Session Target, the harness that runs it: Local, Copilot, Claude, Codex or Cloud. VS Code’s page on agent harnesses↗ (opens in a new tab) says the Local harness offers three roles, Ask, Agent and Plan. The Copilot harness runs on the Agent Host, where Autopilot is an agent mode rather than a permission level, and the Claude and Codex harnesses have their own permission modes. So if you are looking for the word “Agent” and cannot find it, check the Session Target first; you may be in a harness that names things differently.

• Local: Ask, Agent and Plan, with VS Code’s built-in tools, extension tools and your MCP servers.
• Copilot: runs on the Agent Host and can outlive the window. It can currently reach only local MCP servers that need no authentication.
• Claude: turned on or off with github.copilot.chat.claudeAgent.enabled.
• Codex: needs the OpenAI Codex extension or chat.agentHost.codexAgent.enabled.
• Cloud: runs against the repository on GitHub and ends in a pull request, so it sees what is pushed, not what is on your disk.

Which role to use once you have found it is covered in Copilot agent vs ask vs plan.

4. Organisation and enterprise policy

If your Copilot seat comes from an organisation on Copilot Business or Enterprise, its owners decide what you see. GitHub’s guide to Copilot Chat in the IDE↗ (opens in a new tab) says that if you do not see the Agent option in the mode selector, your enterprise or organisation administrator may have disabled agent mode for your IDE, and that you cannot use Copilot Chat at all if they have disabled chat. Policies are hierarchical: an enterprise owner’s choice cannot be overridden by an organisation.

• Agent mode or chat disabled for the IDE: only an owner can change it. Ask them which policy is set, not whether “Copilot is on”.
• MCP servers missing: the “MCP servers in Copilot” policy is disabled by default and applies only to Copilot Business and Enterprise seats. Free, Pro and Pro+ are not governed by it.
• On the VS Code side, administrators can also restrict MCP sources with chat.mcp.access and allow or deny named servers.

5. Agent runs, but a tool or MCP server is missing

Run MCP: List Servers, check each server’s status, open its output and restart it. If a server works in a Local session and is missing in a Copilot session, that is the harness limit above, not a broken server: a remote server that signs in with OAuth is only reachable from Local today. Server configuration itself, servers, inputs and where the file lives, is in VS Code mcp.json.

6. An extension is in the way

Another AI extension, or one that adds its own chat tools, can interfere with chat. Run Help: Start Extension Bisect from the Command Palette. VS Code disables half your extensions and asks whether the problem is still there, and repeats until one is left. If the culprit is an extension you need, keep it but turn off its chat tools; administrators can block third-party extension tools with the ChatAgentExtensionTools policy.

7. AI credits or a rate limit

Premium requests have gone. GitHub’s page on what changed with Copilot billing↗ (opens in a new tab) says that as of 1 June 2026, usage-based billing replaced request-based billing, and every plan now includes a monthly allowance of GitHub AI Credits, used according to the tokens each request consumes. Every agent mode prompt uses them, and a long agent run uses more than a quick question in Ask.

• Check the Copilot status dashboard in VS Code, or your usage dashboard on GitHub, for what is left.
• Once the allowance is used, GitHub’s options are to set a budget for additional usage, upgrade the plan, or wait for the monthly reset.
• A temporary rate limit is separate from the monthly allowance. Wait, start a new session, and try again.

JetBrains, Visual Studio and Xcode

Agent mode exists outside VS Code, but none of the VS Code settings above apply. According to GitHub’s guide to Copilot Chat in the IDE:

• JetBrains IDEs: open the GitHub Copilot Chat icon on the right of the IDE window, then use the Agents dropdown at the bottom of the chat panel. You need the latest version of the GitHub Copilot plugin.
• Visual Studio: View, then GitHub Copilot Chat, then Agent from the agents dropdown. You need Visual Studio 2022 version 17.8 or later; from 17.10 the Copilot extensions are built in.
• Xcode: Editor, then GitHub Copilot, then Open Chat, which opens in its own window, then Agent from the agents dropdown. You need the latest version of the GitHub Copilot extension and to be signed in to GitHub in Xcode.
• In all three, the organisation policies above apply in the same way, and agent prompts use AI credits.

Collect evidence, then report it

If none of this explains it, gather what the VS Code team will ask for before you file anything.

1. Developer: Set Log Level to Trace, reproduce the problem, and read the GitHub Copilot and GitHub Copilot Chat output channels.
2. Run GitHub Copilot: Chat Diagnostics if you suspect the network, a proxy or a VPN.
3. Open the Chat Debug view to see the prompt, context and tools that went to the model for the failing request.
4. Report it with Help, Report Issue, then VS Code Extension, then GitHub Copilot Chat. Include your VS Code version, the steps, what you expected and what happened. Remove secrets and customer data from logs first.

Keep the task where the outage cannot reach it

Most of these faults end a session halfway through, and whatever the agent had decided lives in a chat you may not get back. Keep the task itself on a board. fenbs connects to VS Code as a remote MCP server with an OAuth sign-in, which, according to the harness limit above, means a Local session today. Once connected, the agent can read the task with fenbs_get_item, write its approach into the task’s plan, and comment on what it changed before it stops, so the next session, or the next person, starts from the task rather than from memory. Every change is recorded in History under the assistant’s name. fenbs has no sprints or due dates; a task is simply in To Do, Next Up, In Progress or Completed.