How do AI agents use Cloud Vision API through Open Connector?

Your user connects Cloud Vision API once with one of its cataloged authentication methods. Open Connector stores the credential in an encrypted vault and exposes Cloud Vision API tools to your agent over MCP or a typed API, with credentials injected server-side on each call.

Is this a Cloud Vision API MCP server?

Yes. Open Connector can serve Cloud Vision API as a named MCP server with a scoped allowlist and a per-user connection URL, so any MCP client can call Cloud Vision API actions with credentials injected server-side.

Where do Cloud Vision API credentials live?

In your own infrastructure. Open Connector keeps credentials in its own vault and injects them at call time, so they never leave your environment.

Cloud Vision API integration

Cloud Vision API integration for AI agents.

Cloud Vision API integration for AI agents with secure authentication and server-side credential injection. Open Connector runs the OAuth, seals the token in an encrypted vault, and serves Cloud Vision API tools to your agent over MCP or a typed API — credentials injected server-side, every call audited, nothing leaving your infrastructure. Open source (AGPL-3.0) and self-hostable.

View source on GitHub All integrations →

What your agents can do

Real Cloud Vision API actions, managed and audited.

Your user connects Cloud Vision API once; your agent can then cloud Vision API is a readily available API (REST and RPC) that provides common vision detection features, including image labeling, face and landmark detection, optical character recognition (OCR), and tagging of explicit content — scoped to the OAuth permissions you grant and the tool allowlist you configure. Every action is least-privilege and written to a tamper-evident audit trail.

1
Your user grants Cloud Vision API access once (OAuth) — the token lands in the vault.
2
Your agent calls a tool over MCP or the typed API; Open Connector injects the credential server-side.
3
Every routed call appends a hash-chained audit record — nothing leaves your infra.

Tools & triggers

Supported Cloud Vision API tools.

23 tools are generated from the published Cloud Vision API catalog. Descriptions are plain text; each action remains subject to its configured authentication and tool allowlist.

Showing 23 tools. All published catalog entries are included in this page's server-rendered HTML.

Vision Files Annotate: Service that performs image detection and annotation for a batch of files. Now only "application/pdf", "image/tiff" and "image/gif" are supported. This service will extract at most 5 (customers can specify which 5 in AnnotateFileRequest.pages) frames (gif) or pages (pdf or tiff) from each file provided and perform detection and annotation for each image extracted.; files
Vision Files AsyncBatchAnnotate: Run asynchronous image detection and annotation for a list of generic files, such as PDF files, which may contain multiple pages and multiple images per page. Progress and results can be retrieved through the `google.longrunning.Operations` interface. `Operation.metadata` contains `OperationMetadata` (metadata). `Operation.response` contains `AsyncBatchAnnotateFilesResponse` (results).; files
Vision Images Annotate: Run image detection and annotation for a batch of images.; images
Vision Images AsyncBatchAnnotate: Run asynchronous image detection and annotation for a list of images. Progress and results can be retrieved through the `google.longrunning.Operations` interface. `Operation.metadata` contains `OperationMetadata` (metadata). `Operation.response` contains `AsyncBatchAnnotateImagesResponse` (results). This service will write image annotation outputs to json files in customer GCS bucket, each json file containing BatchAnnotateImagesResponse proto.; images
Vision Locations Operations Get: Gets the latest state of a long-running operation. Clients can use this method to poll the operation result at intervals as recommended by the API service.; locations
Vision Operations Cancel: Starts asynchronous cancellation on a long-running operation. The server makes a best effort to cancel the operation, but success is not guaranteed. If the server doesn't support this method, it returns `google.rpc.Code.UNIMPLEMENTED`. Clients can use Operations.GetOperation or other methods to check whether the cancellation succeeded or whether the operation completed despite cancellation. On successful cancellation, the operation is not deleted; instead, it becomes an operation with an Operation.error value with a google.rpc.Status.code of `1`, corresponding to `Code.CANCELLED`.; operations
Vision Projects Files Annotate: Service that performs image detection and annotation for a batch of files. Now only "application/pdf", "image/tiff" and "image/gif" are supported. This service will extract at most 5 (customers can specify which 5 in AnnotateFileRequest.pages) frames (gif) or pages (pdf or tiff) from each file provided and perform detection and annotation for each image extracted.; projects
Vision Projects Files AsyncBatchAnnotate: Run asynchronous image detection and annotation for a list of generic files, such as PDF files, which may contain multiple pages and multiple images per page. Progress and results can be retrieved through the `google.longrunning.Operations` interface. `Operation.metadata` contains `OperationMetadata` (metadata). `Operation.response` contains `AsyncBatchAnnotateFilesResponse` (results).; projects
Vision Projects Locations Images Annotate: Run image detection and annotation for a batch of images.; projects
Vision Projects Locations Images AsyncBatchAnnotate: Run asynchronous image detection and annotation for a list of images. Progress and results can be retrieved through the `google.longrunning.Operations` interface. `Operation.metadata` contains `OperationMetadata` (metadata). `Operation.response` contains `AsyncBatchAnnotateImagesResponse` (results). This service will write image annotation outputs to json files in customer GCS bucket, each json file containing BatchAnnotateImagesResponse proto.; projects
Vision Projects Locations ProductSets AddProduct: Adds a Product to the specified ProductSet. If the Product is already present, no change is made. One Product can be added to at most 100 ProductSets. Possible errors: * Returns NOT_FOUND if the Product or the ProductSet doesn't exist.; projects
Vision Projects Locations ProductSets Create: Creates and returns a new ProductSet resource. Possible errors: * Returns INVALID_ARGUMENT if display_name is missing, or is longer than 4096 characters.; projects
Vision Projects Locations ProductSets Import: Asynchronous API that imports a list of reference images to specified product sets based on a list of image information. The google.longrunning.Operation API can be used to keep track of the progress and results of the request. `Operation.metadata` contains `BatchOperationMetadata`. (progress) `Operation.response` contains `ImportProductSetsResponse`. (results) The input source of this method is a csv file on Google Cloud Storage. For the format of the csv file please see ImportProductSetsGcsSource.csv_file_uri.; projects
Vision Projects Locations ProductSets List: Lists ProductSets in an unspecified order. Possible errors: * Returns INVALID_ARGUMENT if page_size is greater than 100, or less than 1.; projects
Vision Projects Locations ProductSets Products List: Lists the Products in a ProductSet, in an unspecified order. If the ProductSet does not exist, the products field of the response will be empty. Possible errors: * Returns INVALID_ARGUMENT if page_size is greater than 100 or less than 1.; projects
Vision Projects Locations ProductSets RemoveProduct: Removes a Product from the specified ProductSet.; projects
Vision Projects Locations Products Create: Creates and returns a new product resource. Possible errors: * Returns INVALID_ARGUMENT if display_name is missing or longer than 4096 characters. * Returns INVALID_ARGUMENT if description is longer than 4096 characters. * Returns INVALID_ARGUMENT if product_category is missing or invalid.; projects
Vision Projects Locations Products List: Lists products in an unspecified order. Possible errors: * Returns INVALID_ARGUMENT if page_size is greater than 100 or less than 1.; projects
Vision Projects Locations Products Patch: Makes changes to a Product resource. Only the `display_name`, `description`, and `labels` fields can be updated right now. If labels are updated, the change will not be reflected in queries until the next index time. Possible errors: * Returns NOT_FOUND if the Product does not exist. * Returns INVALID_ARGUMENT if display_name is present in update_mask but is missing from the request or longer than 4096 characters. * Returns INVALID_ARGUMENT if description is present in update_mask but is longer than 4096 characters. * Returns INVALID_ARGUMENT if product_category is present in update_mask.; projects
Vision Projects Locations Products Purge: Asynchronous API to delete all Products in a ProductSet or all Products that are in no ProductSet. If a Product is a member of the specified ProductSet in addition to other ProductSets, the Product will still be deleted. It is recommended to not delete the specified ProductSet until after this operation has completed. It is also recommended to not add any of the Products involved in the batch delete to a new ProductSet while this operation is running because those Products may still end up deleted. It's not possible to undo the PurgeProducts operation. Therefore, it is recommended to keep the csv files used in ImportProductSets (if that was how you originally built the Product Set) before starting PurgeProducts, in case you need to re-import the data after deletion. If the plan is to purge all of the Products from a ProductSet and then re-use the empty ProductSet to re-import new Products into the empty ProductSet, you must wait until the PurgeProducts operation has finished for that ProductSet. The google.longrunning.Operation API can be used to keep track of the progress and results of the request. `Operation.metadata` contains `BatchOperationMetadata`. (progress); projects
Vision Projects Locations Products ReferenceImages Create: Creates and returns a new ReferenceImage resource. The `bounding_poly` field is optional. If `bounding_poly` is not specified, the system will try to detect regions of interest in the image that are compatible with the product_category on the parent product. If it is specified, detection is ALWAYS skipped. The system converts polygons into non-rotated rectangles. Note that the pipeline will resize the image if the image resolution is too large to process (above 50MP). Possible errors: * Returns INVALID_ARGUMENT if the image_uri is missing or longer than 4096 characters. * Returns INVALID_ARGUMENT if the product does not exist. * Returns INVALID_ARGUMENT if bounding_poly is not provided, and nothing compatible with the parent product's product_category is detected. * Returns INVALID_ARGUMENT if bounding_poly contains more than 10 polygons.; projects
Vision Projects Locations Products ReferenceImages Delete: Permanently deletes a reference image. The image metadata will be deleted right away, but search queries against ProductSets containing the image may still work until all related caches are refreshed. The actual image files are not deleted from Google Cloud Storage.; projects
Vision Projects Locations Products ReferenceImages List: Lists reference images. Possible errors: * Returns NOT_FOUND if the parent product does not exist. * Returns INVALID_ARGUMENT if the page_size is greater than 100, or less than 1.; projects

FAQ

Cloud Vision API integration, answered

How do AI agents use Cloud Vision API through Open Connector?: Your user connects Cloud Vision API once with one of its cataloged authentication methods. Open Connector stores the credential in an encrypted vault and exposes Cloud Vision API tools to your agent over MCP or a typed API, with credentials injected server-side on each call.
Is this a Cloud Vision API MCP server?: Yes. Open Connector can serve Cloud Vision API as a named MCP server with a scoped allowlist and a per-user connection URL, so any MCP client can call Cloud Vision API actions with credentials injected server-side.
Where do Cloud Vision API credentials live?: In your own infrastructure. Open Connector keeps credentials in its own vault and injects them at call time, so they never leave your environment.

Give your agents Cloud Vision API — keep the keys.

Open source, self-hostable, with Cloud Vision API credentials that never leave your infrastructure. Run it from source today.

View source on GitHub How the MCP gateway works →