Copilot Coding Agent & Agent Plugins

[filip-be]

Select Topic Area

Question

Copilot Feature Area

Copilot Coding Agent

Body

Hi!

Is there a plan for Copilot Coding Agent to support agent plugins?

I would love to be able to configure Copilot Coding Agent to use some plugin in my repository.
This would allow our organization to distribute common skills/agents/commands/mcp/etc. across different teams and repositories in an easy, opt-in approach.

Thanks for the response on that!

[dhrubojyotihazra]

Yes, this is actually available right now! GitHub Copilot supports exactly what you are describing through a combination of Custom Agents and the Model Context Protocol (MCP).

You don't need to wait for a roadmap; you can configure repository-level (or organization-level) "plugins" and tools today. Here is how it works:

1. Custom Agents
   GitHub Copilot allows you to create specialized Custom Agents tailored to your specific workflows. These are defined using simple Markdown files (.agent.md) that contain YAML frontmatter for configuration and standard Markdown for system instructions.

Through these files, you can define:

Skills/Commands: By providing specific, reusable system instructions.

Tools: Granting the agent access to specific built-in tools (like reading files or terminal execution).

MCP Servers: Extending Copilot's capabilities by connecting it to your own external tools, internal APIs, or custom context sources.

2. Repository-Level Configuration
   To tie an agent (and its MCP tools) to a specific repository, you simply create a .github/agents/ directory in your repo and add your agent profile there (e.g., .github/agents/our-custom-agent.agent.md).

Whenever a developer opens that repository in an IDE like VS Code, or uses Copilot on GitHub.com, they can open Copilot Chat, select the agent dropdown, and choose your custom agent. It is a completely seamless, opt-in approach.

3. Organization-Wide Distribution (Cross-Team)
   Since you mentioned wanting to distribute this across different teams and repositories, you can easily scale this up to the Organization level:

Create a repository named .github-private in your GitHub organization.

Inside that repository, create an agents/ folder at the root (note: just agents/, not .github/agents/).

Place your .agent.md files in that folder.

Once committed, these custom agents will be automatically available across all repositories within your organization. Developers can simply select the agent when they need those specific common skills or MCP integrations.

Example .agent.md Structure
Markdown

name: Org Architecture Expert
description: Connects to our internal APIs to generate boilerplate and fetch schema data.
tools: ['read', 'edit', 'some-mcp-server/tool-1']

You are an expert in our organization's internal architecture.
When asked to build a new feature, always check the internal MCP server for existing schema definitions first...
You can read more about setting this up in the official documentation here:

About custom agents

Extending GitHub Copilot Chat with MCP servers

[adityaprasa231195]

Good news — what you're describing is largely already possible, and it's moved fast. Here's where things stand:
MCP is the plugin mechanism for the Coding Agent today
Repository administrators can configure MCP servers for Copilot Coding Agent directly in repository settings — Settings → Copilot → Coding Agent — using a JSON config. Once configured, the tools from those servers are available to the agent during every assigned task. GitHub
So the opt-in, per-repository model you want is already there. You drop a JSON config into repo settings, point it at your MCP server, and the coding agent picks it up.
The cross-team distribution piece
Organization and enterprise administrators can also configure MCP servers as part of custom agents using YAML frontmatter. GitHub Custom agents can bundle their own MCP configs, and those tools follow a defined processing order: out-of-the-box MCP (like GitHub MCP) first, then custom agent MCP, then repository-level configurations. GitHub
This means you could define a custom agent at the org level that packages your shared capabilities, and teams opt in by using that agent in their repos.
The CLI has a proper plugin system right now
If part of your use case includes Copilot CLI (the terminal agent), plugins are already supported — you can install them directly from GitHub repositories with /plugin install owner/repo. Plugins can bundle MCP servers, agents, skills, and hooks. GitHub This is almost exactly the distribution model you described: host a repo with your organization's shared tooling, teams install it.
The gap: Copilot Extensions + Agent mode
The one thing that doesn't fully work yet is if you've built a Copilot Extension (the GitHub App-based kind). Copilot Extensions currently only work through Ask mode (regular chat), not Agent mode. GitHub has acknowledged this is a platform limitation and noted it's a logical next step, but hasn't committed to a timeline. GitHub
Bottom line for your use case
If you're willing to go the MCP route rather than the Extensions route, the opt-in per-repository distribution story is already there. The practical path today is: build your shared capabilities as MCP servers, host them somewhere your org can reach, and let teams add the JSON config to their repo settings. For CLI-based workflows, the plugin system handles this even more cleanly with a single install command.