Skip to main content
GitHub is Available Now as a source-only connector. MantrixFlow reads selected resources from one GitHub.com repository, stages them with stable relation names, applies published SQL transformations, and delivers the reviewed rows to a supported destination. The data connector is read-only. It does not clone repositories, push commits, open pull requests, change issues, or download release binaries.
The GitHub data source is separate from the GitHub App integration used to version pipeline configuration. Connecting one does not authorize the other.

Supported resources

The source catalog advertises 12 resources: Every root row includes repository identity fields so records remain traceable to the configured repository. A legacy incremental repo_events resource is retained only for previously saved pipelines. It is not offered for new stream selection because GitHub exposes only a limited recent event window and may create tables dynamically by event type. It is not change data capture or a historical audit log.

Before you begin

You need:
  • the GitHub account or organization name that owns the repository;
  • the repository name;
  • a Personal Access Token owned by a user who can read that repository; and
  • permission to create source connections in the MantrixFlow workspace.
For the complete resource set, use a classic Personal Access Token with read access to the repository. Private repositories normally require the classic repo scope. Public-only access can use a narrower token where GitHub allows it. Fine-grained Personal Access Tokens can work for REST resources when their repository access and endpoint permissions are sufficient, but coverage varies by resource. Classic PAT is the supported contract for the GraphQL issue, pull-request, stargazer, comment, and reaction data.
Grant access only to repositories the pipeline needs. MantrixFlow treats the token as an opaque secret, so the token prefix is not used to decide whether it is valid or sufficiently privileged.

Create the source connection

  1. Open Connections in your MantrixFlow workspace.
  2. Click + New Connection and choose Source.
  3. Select GitHub. The card is marked Source only and Available Now.
  4. Enter a descriptive connection name.
  5. Enter the repository owner, such as octocat.
  6. Enter the repository name, such as Hello-World.
  7. Paste the Personal Access Token.
  8. Click Test Connection.
  9. Save after the connection test succeeds.
The connection test validates the authenticated GitHub user and the exact repository. A successful user lookup does not hide a missing repository or insufficient repository permission. The current form accepts one repository per connection. GitHub Enterprise Server and custom API hosts are not supported.

Select and preview resources

  1. Create or open a pipeline that uses the GitHub connection.
  2. Open the Source tab.
  3. Discover the source catalog.
  4. Select the required GitHub resources.
  5. Preview representative records and review the detected fields.
  6. Keep Full Table selected for newly configured resources.
  7. Save the stream configuration.
Preview is read-only and bounded. It does not advance incremental state or modify the repository. An empty resource selection is rejected. An unknown resource name also fails explicitly rather than silently loading a different default resource.

Full Table behavior

All 12 advertised resources use Full Table extraction. Each successful run reads the complete API result visible to the token and replaces the selected staging resource before transformation. Full Table is appropriate for repository snapshots, but these API semantics still matter:
  • rewritten Git history and force pushes can change the commit snapshot;
  • contributor rows are API summaries, not a canonical historical user table;
  • release assets are metadata only and asset files are never downloaded; and
  • the repository events endpoint exposes a limited recent window, commonly around 300 events.

Pagination and rate limits

MantrixFlow follows GitHub REST Link headers and GraphQL opaque cursors until the API reports no next page. Repeated links or cursors fail the run instead of looping forever. The connector applies bounded requests, pages, retries, runtime, and GraphQL query cost. It honors GitHub primary reset headers and Retry-After responses, and uses bounded backoff for secondary limits. A GraphQL response containing errors fails the resource even when GitHub returns HTTP 200; partial data is not reported as complete. Root issue, pull-request, and stargazer pagination is complete. Nested comments, reactions, labels, and assignees are currently limited to the first 100 items per parent.

Transform GitHub data

Use the exact staging relation shown by discovery. For example:
GraphQL nested objects and arrays may be normalized into child relations. Check the discovered schema before referencing comments, reactions, labels, or assignees in a model.

Credential security

  • The PAT is encrypted when saved and masked when returned to the browser.
  • Leaving the masked value unchanged while editing preserves the stored token.
  • Tokens are sent only in authorization headers to fixed HTTPS GitHub.com API endpoints; they are never placed in request URLs.
  • Connector errors and metrics exclude token values and repository content.
  • Rotate or revoke the PAT in GitHub, then re-test the connection in MantrixFlow after updating it.

Current limitations

  • One GitHub.com repository per connection.
  • Personal Access Token authentication only; GitHub App authentication is a separate integration and is not reused by this source.
  • Full Table only for newly selected resources; no advertised CDC.
  • Nested GraphQL connections are limited to 100 rows per parent.
  • Repository discovery exists in the connector runtime, but the current form still uses explicit owner and repository fields.
  • GitHub Enterprise Server, anonymous extraction, conditional requests, and multi-repository extraction are not enabled.
  • Manual UI validation and live source-to-every-destination validation remain pending owner checks; verify destination row counts for each deployment.

Troubleshooting

For connection-wide behavior, see Connections Overview.