Written by: Himanshu Kohli, Product Manager @ Google (Linkedin)
Google Cloud Storage (GCS) is a foundational component of the modern agentic tech stack and the preferred home for unstructured data at scale. But as developers transition from prototyping to deploying autonomous agents in production, a critical architectural shift is happening. The challenge is no longer just storing data; it is turning petabytes of raw, unstructured data into dynamic, agent-accessible context. This is exactly what we aim to solve: GCS is now natively supported and officially available in the MCP Toolbox.
The Rise of Unstructured Data in Agentic Applications
Databases hold structured facts, while object stores hold the reality of a business — PDF contracts, architectural diagrams, financial and compliance documents, raw application logs, and multi-modal media. By standardizing the handshake between agents and GCS, developers are unlocking incredibly powerful applications:
- Multi-modal Analysis: Agents can now combine the deterministic power of structured data from BigQuery and AlloyDB with the rich context of unstructured data on GCS. For example, an automated insurance claim agent can pull a user’s structured policy limits and deductibles from an AlloyDB database, while seamlessly retrieving unstructured crash scene photos and scanned police reports from GCS to assess damage and recommend a payout.
- Automated Document Pipelines: Background agents that monitor specific GCS “intake” buckets for, let’s say, newly uploaded vendor invoices or legal contracts, automatically extracting key entities, and triggering downstream ERP workflows.
- Codebase and Log Forensics: Site Reliability Agents (SREs) can seamlessly dive into buckets containing massive historical server logs or crash dumps to perform root-cause analysis without a developer having to manually download gigabytes of text.
- Dynamic “On-the-Fly” RAG: This is an interesting new pattern we are seeing where customers are moving beyond rigid vector-embedding pipelines. Instead of chunking and indexing everything upfront, you can grant an agent secure access to a specific GCS directory. When a user asks a complex question, the agent can dynamically list the bucket contents, locate the relevant 200-page manual, read the raw file, and synthesize the answer directly from the source.
For more such interesting stories, see this blog containing details on how Snap, Palo Alto Networks, and AirWallex are using GCS with MCP.
You Asked, We Delivered: GCS Support is Here
Since the launch of the MCP Toolbox, native support for massive object stores like GCS has been one of the most highly requested features by our customers. We heard your feedback, and we delivered. Like our other Toolbox sources, the GCS integration is super easy to use. To get your clients connected immediately, follow the quick setup instructions explained here.
GCS’s Toolbox integration comes loaded with pre-built tools that cover most of the core GCS functionalities for creating, listing, reading, and updating buckets, objects, and their metadata. To make management easier, we also provide two built-in toolsets:
- cloud-storage-buckets: Equips your agent with the tools to manage storage infrastructure (e.g., list buckets, get bucket metadata).
- cloud-storage-objects: Grants your agent the granular ability to interact with the unstructured data itself (e.g., list files, read object contents, upload, and delete).
Stay tuned for additional new interesting tools coming up in the next few months!
Using Custom Tools for Prescriptive, Advanced Business Logic
While for most of the use-cases pre-built tools are sufficient, we are seeing a number of patterns that require customers to incorporate custom business logic. The good news is, MCP Toolbox already provides a declarative, extensible framework — custom tools. You can use the native GCS integration as a foundational building block to create custom tools tailored to your specific business needs. For example: when you want to restrict an agent’s access to only a few specific buckets without risking any future leaks. By wrapping the GCS integration into a custom tool, you can hardcode an allow-list, entirely eliminating the risk of prompt injection leading to cross-bucket traversal. Below is an example:
kind: tool
name: list_allowlisted_buckets
type: cloud-storage-list-buckets
source: my-gcs-source
description: Use this tool to list Cloud Storage buckets allowlisted for the vendor risk assement agent
project: my_safe_project" # hard-coded GCP project
prefix: ai_buckets # allowlisted bucket prefix
Similarly, allow reading objects from only an allowlisted, safe bucket.
kind: tool
name: read_allowlisted_objects
type: cloud-storage-read-object
source: my-gcs-source
description: Use this tool to read contents of available, allowlisted Cloud Storage objects
bucket: bucket_for_agents # allowlisted GCS bucket
Get Started
By connecting your largest unstructured data stores directly to your AI agents, the possibilities
are virtually limitless. We are incredibly excited to see what you build with GCS and the MCP Toolbox.
- Visit the official documentation to quickly get started with GCS and Toolbox.
- Learn more about the GCS Pre-built tools and toolsets.
- Explore the source code on the MCP Toolbox GitHub repository (don’t forget to drop a star! ⭐)
Turning Data into Context: Google Cloud Storage (GCS) is Now Available in MCP Toolbox was originally published in Google Cloud – Community on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source Credit: https://medium.com/google-cloud/turning-data-into-context-google-cloud-storage-gcs-is-now-available-in-mcp-toolbox-5880f368846a?source=rss—-e52cf94d98af—4
About the Author
