Loading ToolBench

Reading the current corpus…

gcs-mcp-server MCP server · D · ToolBench

← Browse

gcs-mcp-server

github.com/HappyPathway/gcs-mcp-server

MCP server for Google Cloud Storage operations

<=2025-11-25STDIO9 tools

51/100

Definition Quality69

Protocol Readiness47

Supportability23

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

GCS MCP Server has clear, action-verb naming (list_, create_, delete_, get_, upload_, copy_) and reasonable descriptions (avg ~80 chars). All 9 tools have input schemas with typed parameters and descriptions. However, schemas lack enums for constrained values (e.g., storage_class, location), parameter descriptions are minimal (10-30 chars), and output schemas are not formally documented. Error handling is absent, no guidance on retryability, user-fixable errors, or recovery paths. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Composition is sound (single responsibility per tool), but missing pagination for list operations and no confirmation pattern for destructive ops.

Tools (9)

copy_objectwriteauthsource verified68/100

Copy an object from one location to another in GCS

create_bucketwriteauthsource verified68/100

Create a new GCS bucket

delete_bucketdestructiveauthsource verified67/100

Delete a GCS bucket

delete_objectdestructiveauthsource verified67/100

Delete an object from GCS

get_bucket_objectsread onlyauthsource verified68/100

List objects in a GCS bucket with optional prefix filter

list_bucketsread onlyauthsource verified75/100

List all GCS buckets in the project with their details

list_object_versionsread onlyauthsource verified65/100

List all versions of an object if versioning is enabled

Issues (5)

HIGHconstrained-input ↗

No enum constraints for storage_class (STANDARD, NEARLINE, COLDLINE, ARCHIVE) or location parameters. LLMs will hallucinate invalid values.

create_bucket

HIGHtool ↗

Output schemas not formally documented. Tool descriptions state what is returned (e.g., 'returns list of bucket info') but JSON Schema output type is missing. LLMs cannot plan downstream calls without knowing response structure.

list_bucketscreate_bucketget_bucket_objectsupload_object

HIGHconfirmation-request ↗

No error handling or recovery guidance. Destructive operations (delete_bucket, delete_object) lack confirmation pattern or dry-run support. No error messages guide LLM on retryability or user-fixable issues.

delete_bucketdelete_object

MEDIUMtool-annotation ↗

Missing tool annotations. Tools marked with risk levels (READ_ONLY, WRITE, DESTRUCTIVE) in metadata but not exposed via readOnlyHint/destructiveHint in schema. LLMs cannot infer safety properties.

Recommendations

Add enum constraints to storage_class (STANDARD, NEARLINE, COLDLINE, ARCHIVE) and location (US, EU, ASIA_EAST1, etc.) parameters in create_bucket. Replace free-form strings with machine-parseable enums.
Document output schemas formally. For each tool, add a 'returns' section specifying the JSON structure: e.g., 'Returns: {name: string, created: ISO8601 string, location: string, storage_class: string}'.
Implement pagination for list_buckets, get_bucket_objects, and list_object_versions. Add limit (default 20, max 100) and offset/cursor parameters. Return total_count or next_cursor to enable multi-page iteration.
Add tool annotations via readOnlyHint and destructiveHint in tool schemas. Mark read-only tools (list_*, read_*, get_*) with readOnlyHint=true; mark destructive tools (delete_*) with destructiveHint=true.
Implement error handling with recovery guidance. For delete_bucket with force=false on non-empty bucket, return 'Bucket not empty. Set force=true to delete all objects first, or call get_bucket_objects to review contents.' For invalid storage_class, return 'Invalid storage_class: got "INVALID", must be one of: STANDARD, NEARLINE, COLDLINE, ARCHIVE'.

Spec posture evidence

Inferred effective spec: <=2025-11-25.

Relies on Logging (deprecated) - log to stderr or use OpenTelemetry

Score history

Overall score trend

First recorded score · v2 rubric

51/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	51	<=2025-11-25	v2