Connect GitHub to Codex

Review open GitHub pull requests in Codex. Connect your account, choose a repository, and retrieve up to five open PRs through OOMOL MCP.

Start Working with GitHub

Start Working with GitHub

Authorize your GitHub account through OAuth.

Check Resource Access First

A GitHub account with access to the repositories you want to use.

Connect GitHub

Sign in to OOMOL, select your team, and create the GitHub connection using a supported authorization method. Check that the connection is active, then use the same team in your MCP configuration or oo CLI login.

Connect GitHub
  • A GitHub account with access to the repositories you want to use.
  • Repository authorization uses the repo scope, which includes read and write access. Available operations also depend on your account’s repository permissions.
Choose Your Access Method

For clients that support remote HTTP MCP and custom authentication headers.

  1. Open the MCP setup page, sign in, and select the team containing your application connection. Use Copy MCP Configuration to get the key and headers.
  2. Add the configuration using the MCP client instructions below, then reconnect your client.
  3. Copy the read-only verification task into your AI chat and run it.
Configure MCP MCP Client Configuration →

For agents that can run terminal commands and use the oo Skill.

  1. Follow the agent setup page to install oo CLI and the oo Skill.
  2. Sign in to the OOMOL team containing your application connection. Use oo login --team <team-name> with your team name.
  3. Copy the read-only verification task into your agent chat and run it.
Set Up oo CLI + Skill CLI Installation Guide →

MCP Client Configuration

Get MCP Configuration

Replace the placeholders with the key and team name from the OOMOL setup page. Keep credentials private and preserve other server entries.

Codex

Open ~/.codex/config.toml and add or update the single oomol entry below. Replace <team-name> with the team in your copied OOMOL configuration. Set OOMOL_MCP_API_KEY in the environment of the process that launches Codex, using only the key from Authorization, without the Bearer prefix. Remove any old Authorization entry from oomol.http_headers and keep the key out of config.toml. Restart Codex with that environment.

[mcp_servers.oomol]
url = "https://connector.oomol.com/v1/mcp"
bearer_token_env_var = "OOMOL_MCP_API_KEY"

[mcp_servers.oomol.http_headers]
"x-oo-team-name" = "<team-name>"

For Codex CLI on macOS or Linux with Bash, run this command and paste the key at the hidden prompt. It passes the key to this Codex process without writing it to a file or shell history. A desktop or IDE client must receive the same variable in its own launch environment; a variable set in another terminal may not be available.

bash -c '
read -r -s -p "OOMOL MCP API key: " OOMOL_MCP_API_KEY || exit 1
export OOMOL_MCP_API_KEY
exec codex
'
Verify the Connection
  1. In Codex CLI, use /mcp to check oomol. In a desktop or IDE client, check its MCP status. Confirm the process receives OOMOL_MCP_API_KEY, then start a new chat.
  2. Run the read-only verification task in a new AI chat to check access to your resource and its data.

Look for OOMOL tools such as connector_list_apps and connector_search_actions. They discover and run GitHub actions; each action may not appear as a separate MCP tool. Frequently Asked Questions →

Verify Your First GitHub Query

Copy this read-only task into a new AI chat. If it fails, check the FAQ.

Choose Your Access Method

Verify Your First GitHub Query

Check that the returned resource identifier matches your target and review the query results.

Use OOMOL MCP only. Check that GitHub has an active connection in the configured OOMOL team. Read the connected GitHub user, list accessible repositories, and let me select one. Read at most 5 open pull requests. Return the repository, PR numbers, titles, URLs, and draft flags when available. Do not infer review approval or passing checks. Ask me to choose when the target is unclear. Report only actual results, identify the selected resource, and distinguish empty results from access errors. Do not create, update, delete, or send anything.

Frequently Asked Questions

Why can I connect but not read my resource?
Use the OOMOL team containing your active application connection. Application authorization, the MCP key, and resource permissions are separate checks. Inspect the tool error and confirm the resource ID and query range. Report empty results separately from access errors; keep keys private. For GitHub, check the connected user and repository access. The repo scope includes write access even though this first task only reads. A PR listing does not verify reviews, CI checks, or permission to merge.
What can AI help me find when triaging issues?
Ask it to group related reports, summarize comments, and identify missing reproduction steps. Specify a repository and the issue criteria; prioritization should follow the impact and urgency information you provide.
Can AI use private repositories?
Private repository access depends on the granted repo scope and the connected account’s access. A connection does not bypass repository permissions.
What should I provide for AI to analyze a pull request?
Provide the pull request URL and your review focus. The assistant can summarize changed files and inspect the returned diffs for potential risks; ask it to flag missing context rather than treating the analysis as a complete review.
Does analyzing a pull request submit a review or merge it?
Reading a pull request and its files does not submit a review or merge it. Those are separate supported operations and should be requested explicitly.