Run publishing checks in CI/CD
Check every pull request and deployment of your MCP server with the publishing checklist, through Manufact's GitHub reporting or the Manufact Publishing Checks GitHub Action.
You can run the publishing checklist on every change to your MCP server, so a pull request shows whether it is still ready for ChatGPT and Claude before you merge it. There are two ways to do it:
- GitHub reporting (no workflow file). For Manufact-hosted servers with a connected GitHub repository, Manufact runs the checklist after each preview deployment and reports the result on the pull request.
- The Manufact Publishing Checks GitHub Action. Add a step to any workflow. Use it to fail the job on errors or warnings, use the results in later steps, check production on push, or check servers you host elsewhere.
Which one do I need?
| You want to | Use |
|---|---|
| See checklist results on pull requests for a Manufact-hosted server | GitHub reporting |
| Fail a workflow job when checks fail, or when they report warnings | GitHub Action |
| Use the score or client readiness in later workflow steps | GitHub Action |
Check production after each push to main, or on a schedule | GitHub Action |
| Check preview deployments of a server you host elsewhere (for example on Vercel) | GitHub Action |
Pick one per pull request
Each checklist run is billed as a normal checklist run. If you turn on GitHub reporting for preview branches and also run the action on pull_request, every push runs the checklist twice and posts two comments.
Report checks on preview deployments with GitHub reporting
When a Manufact-hosted server is connected to a GitHub repository, pushed branches deploy as preview environments (the branch allow list in Server settings controls which branches deploy). GitHub reporting runs the checklist once each preview deployment is live, then reports the result on the commit and its pull request. You don't need a workflow file, an API key, or a repository secret.
GitHub reporting is off by default. To turn it on:
- Open your server and go to Settings → Checklist automation.
- Turn on Run automatically and choose All deployed branches, or Selected branches with patterns that match your pull request branches. Production branch only does not run on preview deployments.
- Turn on Report publishing checks to GitHub.
For each preview deployment, Manufact then:
- adds a Manufact - Publishing checks check to the deployed commit, with the result Success, Failure, or Neutral;
- creates or updates one Publishing checks comment on the open pull request, with tables for required fixes, policy concerns, and recommendations;
- links both back to the full checklist run in the dashboard.
A failing or unfinished checklist never marks the deployment itself as failed. See Report results to GitHub for what each result means.
Run checks with the GitHub Action
The mcp-use/publishing-checks action runs the checklist from a workflow job, prints the report to the job log and job summary, and comments it on the pull request. The job fails when the checklist does not pass the level you set in fail-on.
Before you start
- Create an API key in the organization that owns the server.
- In your GitHub repository, go to Settings → Secrets and variables → Actions and add the key as a secret named
MANUFACT_TOKEN. - Find your server ID. In the examples below, replace
<server-id>with it.
Find your server ID
The server ID is a UUID that identifies your server in Manufact. It is not a secret, so you can put it in the workflow file directly.
In the dashboard: open the server. The server ID is the UUID after /servers/ in the browser's address bar:
https://manufact.com/cloud/<org>/servers/e3b0c442-98fc-4c14-9afb-f4c8996fb924/overview
└─────────── server ID ────────────┘With the CLI: run mcp-use servers list. It prints each server's name, status, and full ID.
Check each pull request
This workflow checks the preview deployment of the pull request's head branch and fails the job when a publishing check fails:
name: Publishing checks
on:
pull_request:
permissions:
pull-requests: write # to comment the report on the pull request
jobs:
checklist:
runs-on: ubuntu-latest
steps:
- uses: mcp-use/publishing-checks@v1
with:
token: ${{ secrets.MANUFACT_TOKEN }}
server: <server-id>The action checks whatever is live on the branch's preview environment when the step runs. It does not wait for a deployment that is still building. Run it after the branch has deployed: if your deploy runs in the same workflow, put this step after it.
Manufact-hosted pull requests
When Manufact deploys your branches from GitHub, the deployment and a pull_request workflow start on the same push, so the action can check the previous deployment. To check each preview once it is live, use GitHub reporting instead.
Check production after each push
On events other than pull_request, the action checks production. This workflow checks production after every push to main:
name: Publishing checks (production)
on:
push:
branches: [main]
jobs:
checklist:
runs-on: ubuntu-latest
steps:
- uses: mcp-use/publishing-checks@v1
with:
token: ${{ secrets.MANUFACT_TOKEN }}
server: <server-id>To always check production, including on pull requests, set branch: "".
Fail on warnings
By default the job fails only when a publishing check fails. Set fail-on: warnings to also fail on warnings and policy concerns:
- uses: mcp-use/publishing-checks@v1
with:
token: ${{ secrets.MANUFACT_TOKEN }}
server: <server-id>
fail-on: warningsfail-on | The job fails when |
|---|---|
none | The checklist can't run at all, for example because the API key is wrong |
failed (default) | A publishing check fails, or the run itself fails or is cancelled |
warnings | Same as failed, or a check reports a warning or a policy concern |
Policy concerns count as warnings, as they do in the dashboard: they may need store review, but they are not failed publishing checks.
Use the results in later steps
Set fail-on: none to keep the job green and decide what to do with the outputs yourself:
- id: checklist
uses: mcp-use/publishing-checks@v1
with:
token: ${{ secrets.MANUFACT_TOKEN }}
server: <server-id>
fail-on: none
- if: steps.checklist.outputs.ready-for-chatgpt == 'true'
run: echo "Ready for ChatGPT (score ${{ steps.checklist.outputs.score }})"Check preview deployments of a server you host elsewhere
If your server runs on another host and is connected to Manufact by URL, run the checklist after each preview deployment and pass that deployment's MCP URL in url:
name: Publishing checks
on:
deployment_status:
permissions:
pull-requests: write # to comment the report on the pull request
jobs:
checklist:
# Check finished preview deployments only.
if: github.event.deployment_status.state == 'success' && github.event.deployment.environment == 'Preview'
runs-on: ubuntu-latest
steps:
- uses: mcp-use/publishing-checks@v1
with:
token: ${{ secrets.MANUFACT_TOKEN }}
server: <server-id>
url: ${{ github.event.deployment_status.environment_url }}/mcp
profile: <profile-id> # optionalurlis the MCP endpoint to check. The run is saved under the server, and the report shows which URL and commit it checked.profileis the ID of the profile to sign in with when the server needs OAuth. Leave it out to use the server's own profile. The profile must be yours or shared with the organization.deployment_statusevents fire only when your host creates GitHub deployments. On Vercel, check that creating GitHub deployments is turned on for the project.- The action finds the open pull request for the deployed commit and comments there.
- Your preview deployments must accept the profile's token: they need to trust the same authorization server as the profile, and accept a token issued for the MCP resource the profile signed in to.
Pull request comment
On pull_request events, and on deployment_status events for a commit with an open pull request, the action comments the full report on the pull request. The comment follows the dashboard's checklist layout: the score, client readiness, the policy review banner, and one section per category, with failing checks and how to fix them.
- The action keeps one comment per server and updates it on later runs.
- If the checklist can't run at all, the comment shows the error instead.
- Commenting needs
pull-requests: write. Without it, the action logs a warning and the checklist result is unaffected. - Pull requests from forks get a read-only token, so they get no comment.
- Set
comment: falseto turn comments off. The report still goes to the job log and the job summary.
Inputs
| Input | Default | Description |
|---|---|---|
token | Required | Manufact API key for the server's organization. Store it as a secret. |
server | Required | Server ID. See Find your server ID. |
branch | The pull request's head branch; production on other events | Branch whose deployment is checked. Set branch: "" to always check production. Ignored when url or profile is set. |
url | None | Servers connected by URL only. The MCP URL to check instead of the server's, for example a preview deployment. |
profile | None | Servers connected by URL only. ID of the profile to sign in with instead of the server's. |
fail-on | failed | none, failed, or warnings. See Fail on warnings. |
comment | true | On pull requests, comment the full report. Set false to turn it off. |
github-token | github.token | Token used to post the comment. |
Outputs
| Output | Description |
|---|---|
run-id | Checklist run ID |
status | Final run status, for example completed or failed |
score | Checks score out of 100, as shown in the dashboard |
ready-for-chatgpt | true or false; empty when unknown |
ready-for-claude | true or false; empty when unknown |
dashboard-url | Link to the checklist run in the dashboard |