# GitHub Copilot cloud agent: features, protocols, quickstart

## TL;DR

GitHub Copilot cloud agent is GitHub's hosted coding agent. You assign it an issue or prompt, and it works in the background in an ephemeral GitHub Actions environment, then opens a pull request. Several sessions can run at once, and Claude and Codex agents can be chosen in preview. It suits teams that already manage work in GitHub.

## Key facts

| Field | Value |
| --- | --- |
| Type | Platform |
| Languages / SDKs | Not applicable (hosted product) |
| License | Proprietary |
| Pricing model | [Hosted, paid](https://docs.github.com/en/copilot/get-started/plans) |
| Orchestration pattern | Other |
| Repository | No public repository (closed source) |
| Website | [github.com](https://github.com/features/copilot) |
| Documentation | [docs.github.com](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent) |
| Last verified | 2026-09-30 |

## Key features

- Tasks start from an issue assignee, a prompt in the Agents tab or panel, a pull request `@copilot` comment, VS Code, Slack, Microsoft Teams, Jira, Linear or Azure Boards, and the agent works on a branch before opening a pull request. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent))
- The Agents tab lists sessions per repository, shows live session logs, allows steering a running session, and can run multiple agent sessions concurrently. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/agent-management))
- Custom agents are Markdown agent profiles with a prompt, tools and MCP servers, so a repository can define task-specific agents such as a test writer or a documentation agent. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-custom-agents))
- Third-party coding agents (Anthropic Claude and OpenAI Codex, public preview) can be assigned issues and pull requests alongside Copilot, and partners can ship agent apps as GitHub Apps. ([source](https://docs.github.com/en/copilot/concepts/agents/about-third-party-coding-agents))
- Automations run the cloud agent on a schedule or when a repository event fires, for example to triage new issues or fix failing tests nightly. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-automations))
- An agent tasks REST API (public preview) and the `gh agent-task` commands start and track cloud agent sessions programmatically. ([source](https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/use-cloud-agent-via-the-api))
- Commits are authored by Copilot with the requester as co-author, are signed, and link back to the session log for review. ([source](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/manage-and-track-agents))

## Architecture and orchestration pattern

Pattern: Other.

The cloud agent is a hosted service inside GitHub. For each task it starts an ephemeral development environment powered by GitHub Actions, where it can explore the code, edit files on a branch, and run tests and linters. Commits, the session log and the pull request are the record of what it did. Only one repository and one branch are in play per task, and a task yields at most one pull request.

Multi-agent use comes from running sessions side by side and from choosing among agents. The docs say you can run multiple agent sessions concurrently, and the getting-started guide has you start a coding task and a research task in parallel. Each session is independent: there is no described coordinator that splits work among sessions or passes messages between them. Custom agents (Markdown profiles), agent apps from partners and third-party agents (Claude, Codex) are alternative workers you assign a task to.

Context is limited to the target repository by default, extended through repository MCP settings, custom instructions, skills, and Copilot Memory (public preview on some plans). A chat started in Copilot Chat on GitHub hands its context to the cloud session it launches.

### Human in the loop

Assigning the task is the human step, and the output is a pull request that a person reviews. You can steer a running session with follow-up messages (each costs AI credits), stop it, and archive it. Workflows do not run automatically on Copilot's pushes; a person clicks Approve and run workflows unless the repository setting allows it. If a repository requires approvals, the approval of the person who started the Copilot pull request does not count. Copilot uses configured MCP tools without asking for approval, so the docs advise limiting the tools list.

### Harnesses it can drive

- Codex ([evidence](https://docs.github.com/en/copilot/concepts/agents/openai-codex))

## Protocols

| Protocol | Support | Evidence | Note |
| --- | --- | --- | --- |
| MCP | Yes (checked 2026-09-30) | [link](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/mcp-and-cloud-agent) | Client: the cloud agent uses tools from local and remote MCP servers, with GitHub and Playwright servers enabled by default. It supports MCP tools only (not resources or prompts) and not remote servers that require OAuth. |
| A2A | Unknown (checked 2026-09-30) | — | No mention of A2A or Agent2Agent in the cloud agent, third-party agent, agent app, automation and API docs pages read. Closed-source, so no repository to search. |
| AG-UI | Unknown (checked 2026-09-30) | — | No mention of AG-UI in the same docs pages, and the product is not listed in the AG-UI protocol README. |

## Best for

- Teams that track work in GitHub Issues and want routine tickets (bug fixes, tests, docs, tech debt) handed to a background agent that returns a pull request. ([shortlist](https://multiagentguide.top/best/coding-agents.md))
- Running several independent agent sessions on different tasks or repositories from one Agents panel, with logs and steering. ([shortlist](https://multiagentguide.top/best/parallel-coding-agents.md))
- Organisations that want agent work governed through GitHub policies, repository rulesets, session logs and usage metrics. ([shortlist](https://multiagentguide.top/best/enterprise-governance.md))

## Not for

- Repositories hosted outside GitHub: the docs say the agent only works with repositories hosted on GitHub.
- Tasks that span several repositories in one run or need more than 59 minutes; the docs give both as limits.
- Users who need agents to coordinate or hand work to each other automatically; sessions are independent.

## Quickstart

```sh
Open https://github.com/copilot/agents with a paid Copilot plan and cloud agent enabled for the repository (Business and Enterprise need an administrator to enable it).
```

Install not yet verified by this site.

```text
# In the GitHub web UI (from the official getting-started guide)
1. Open an issue in a repository where Copilot cloud agent is enabled.
2. Assignees -> Copilot -> optional prompt -> Assign.
3. Open the agents panel and start a second task, for example:
   Investigate which dependencies in this repo are outdated and summarize what upgrading would involve.
4. Open either session to read the log; type in the prompt box to steer.
5. When a pull request appears, review it; Approve and run workflows if you are comfortable.

# From a terminal (GitHub CLI v2.80.0 or later, public preview)
gh agent-task create "Fix the login button on the homepage" --follow
gh agent-task list
```

### Common pitfalls

- Available on paid Copilot plans only; Business and Enterprise need an administrator policy, and repository owners can opt repositories out.
- One repository, one branch and one pull request per task; sessions stop at 59 minutes (a hard limit).
- Rulesets or branch protection that limit commit authors can block the agent; add Copilot as a bypass actor if needed.
- The agent sees the issue text and comments as they were at assignment; put later information on the pull request.
- Repository MCP settings support tools only and not OAuth-protected remote servers.
- The agent tasks API and `gh agent-task` are public preview; the API accepts user-to-server tokens only.
- Steering messages and sessions consume GitHub AI Credits and Actions minutes.

Official quickstart: https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/overview

## Pros

- Runs inside GitHub: work starts from issues, pull requests, Slack or Teams, and results come back as commits and a pull request with a linked session log. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent))
- Sessions can run concurrently, be watched live and be steered without stopping. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/agent-management))
- Lets you pick among Copilot, custom agents, partner agent apps and third-party agents (Claude, Codex) for a task. ([source](https://docs.github.com/en/copilot/concepts/agents/about-third-party-coding-agents))
- Safe defaults around workflows: Actions do not run automatically on the agent's pushes, and the requester's own approval does not count toward required reviews. ([source](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/review-copilot-output))
- A REST API and GitHub CLI commands let other tools start and track tasks. ([source](https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/use-cloud-agent-from-cli))

## Cons

- One repository, one branch at a time and exactly one pull request per task; cross-repository changes in one run are not supported. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent))
- Each session has a hard 59-minute maximum, and the agent only works with repositories hosted on GitHub. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent))
- MCP support covers tools only and excludes OAuth-protected remote servers; Copilot uses configured tools without asking for approval. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/mcp-and-cloud-agent))
- Third-party agents, agent apps, the agent tasks API and the `gh agent-task` commands are public preview, and steering is not available for third-party coding agents. ([source](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/manage-and-track-agents))
- Usage is metered: sessions consume GitHub Actions minutes and AI credits, and steering messages cost credits. ([source](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent))

## Alternatives

- [Cursor](https://multiagentguide.top/tools/cursor.md)
- [Devin](https://multiagentguide.top/tools/devin.md)
- [Tembo](https://multiagentguide.top/tools/tembo.md)
- [OpenHands](https://multiagentguide.top/tools/openhands.md)
- [Codex CLI](https://multiagentguide.top/tools/codex.md)

## FAQ

### Does the Copilot cloud agent support MCP?

Yes, as a client for MCP tools. The GitHub and Playwright MCP servers are on by default, and repository administrators can add more. Resources, prompts and OAuth-protected remote servers are not supported.

### Can several agents work at the same time?

Yes. The docs say multiple agent sessions can run concurrently, and each session works on its own task and branch. There is no documented coordinator that splits one task across sessions.

### What does it cost?

It is included with paid Copilot plans. The plans docs page lists Pro at $10 per month, Pro+ at $39, Max at $100, Business at $19 per seat and Enterprise at $39 per seat (read 2026-09-30). Sessions use GitHub Actions minutes and AI credits.

### Is it the same as agent mode in the IDE?

No. Agent mode edits your local files in the IDE; the cloud agent works autonomously in a GitHub Actions environment and delivers a branch or pull request.

### Does it run Claude Code?

The docs describe an Anthropic Claude coding agent (public preview) that uses the Claude Agent SDK and an OpenAI Codex agent that uses the Codex SDK. They do not name Claude Code, so only Codex is listed under harnesses.

## Sources

- [GitHub Copilot features](https://github.com/features/copilot)
- [About GitHub Copilot cloud agent](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent)
- [About agent management](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/agent-management)
- [About custom agents](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-custom-agents)
- [About third-party coding agents](https://docs.github.com/en/copilot/concepts/agents/about-third-party-coding-agents)
- [OpenAI Codex coding agent in Copilot](https://docs.github.com/en/copilot/concepts/agents/openai-codex)
- [About Copilot automations](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-automations)
- [Using Copilot cloud agent via the API](https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/use-cloud-agent-via-the-api)
- [Using Copilot cloud agent from the GitHub CLI](https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/use-cloud-agent-from-cli)
- [Managing agent sessions](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/manage-and-track-agents)
- [Review output from Copilot](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/review-copilot-output)
- [Model Context Protocol and Copilot cloud agent](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/mcp-and-cloud-agent)
- [Plans for GitHub Copilot](https://docs.github.com/en/copilot/get-started/plans)
- [Get started with Copilot agents on GitHub](https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/overview)
- [GitHub Docs llms.txt](https://docs.github.com/llms.txt)

## Unknown fields

A2A and AG-UI: searched the cloud agent, agent management, third-party agent, agent app, automation, MCP and API docs pages for A2A, Agent2Agent and AG-UI and found nothing; the product is closed-source so no repository search was possible. Pricing: plan prices are from the docs plans page (https://docs.github.com/en/copilot/get-started/plans); the github.com/features/copilot/plans page was not used for numbers because it lists additional figures whose meaning was not clear from the page text. Orchestration pattern 'other': sessions are independent and no coordinator pattern is documented. The multi-agent function here is parallel sessions plus a choice of custom, partner and third-party agents.

Corrections or removal requests: support@multiagentguide.top

---

Data as of 2026-09-30. Not affiliated with listed projects. HTML version: https://multiagentguide.top/tools/github-copilot-cloud-agent