Start nowCreate an accountSign in
Start now
Documentation

Search documentation

Try deployment, Codex, or data access.

    Browse documentation

    Install the SpringRoll skill in Codex and connect its tools

    A skill gives Codex a repeatable workflow; an MCP connection supplies the tools that carry it out. Install the SpringRoll skill in your project, connect its matching endpoint and inspect your workspace before deploying. This GitHub installation works independently of a public plugin-directory listing.

    By Updated

    On this page

    Choose a skill, a plugin or a manual connection

    OpenAI plugins can bundle skills and MCP configuration. Installing a skill alone does not register the MCP server or authorize a workspace. A public GitHub repository, a local plugin source and a universal directory listing are separate distribution channels.

    Scroll sideways to see all columns. With the table focused, use the arrow keys.

    Choose the setup that matches the instructions you use.
    RouteWhat you getSpringRoll endpoint
    Skill from GitHubProject instructions and bundled reference files; connect MCP separately/api/mcp/chatgpt, matching the current skill’s proposal and confirmation tools
    Plugin packageThe skill and the package’s MCP configurationRead the installed package’s endpoint; the verified repository uses /api/mcp/chatgpt
    Manual coding-client setupGeneral MCP tools without the packaged skill workflow/api/mcp; follow the separate Codex MCP guide

    1. Install the skill in your project

    From the project folder, run the command below. It selects Codex and copies the skill into the project’s .agents/skills/springroll directory. It does not install globally. Review the downloaded SKILL.md and start a new Codex session after installation.

    npx skills@1.7.6 add springroll-dev/springroll-skill --skill springroll --agent codex --copy

    2. Add the endpoint used by this skill

    The public skill currently calls proposal and confirmation tools such as springroll_prepare_deploy. Those belong to /api/mcp/chatgpt. The endpoint name does not limit it to the ChatGPT interface; use this matching URL when setting up the skill in Codex.

    Check codex mcp list first. If you already have a SpringRoll connection, keep it: a saved credential identifies its workspace. The distinct springroll-skill name below avoids overwriting an existing manual connection. A new connection is not guaranteed to reuse an earlier guest workspace.

    codex mcp list
codex mcp add springroll-skill --url https://springroll.dev/api/mcp/chatgpt
codex mcp login springroll-skill

    3. Start as a guest and inspect the workspace

    Review the requested permissions in the browser and choose Connect SpringRoll. You can start without creating a SpringRoll account or signing in; an existing signed-in member can use their authorized workspace. Then start a new Codex session in the project.

    Ask Codex to use the SpringRoll skill, read the connected profile, workspace, company guidance and supported workloads, and report the owner and any blockers before changing anything. Confirm that the skill is available and that its requested tools exist. Missing tools usually mean the connection and skill use different endpoints.

    Use the SpringRoll skill. Inspect this project, show the connected workspace and permissions, read its company guidance and list supported deployment workloads. Do not deploy yet.

    4. Review a deployment before confirming it

    Ask Codex to prepare a deployment for your own project or an example with fictional records. Review the proposed target, build settings and hosting estimate. Confirm that specific proposal when ready, then follow the operation to its actual result and test the app.

    The skill does not grant company-data access or waive a required Staged review. Choose the App Portal audience and protect the direct provider URL separately. Keep the saved connection to return to the guest workspace; creating a separate account does not transfer guest apps automatically.

    What we verified, and what still depends on your account

    On 8 November 2026, skills 1.7.6 installed the springroll skill into a new project directory successfully. We inspected the public repository at revision d50db28667380413759c0b08b7d6d1600004a8b9 and its package configuration. The local Codex CLI was 0.151.0. This verifies the installer and package contents, not a deployment in your workspace.

    We have not verified a public SpringRoll listing in OpenAI’s universal plugin directory. Use the GitHub skill route above today. If your organization distributes the plugin privately, follow its approved installation instructions and inspect the bundled endpoint. Directory publication is not a prerequisite for the GitHub skill route.