An implementation of a Model Context Protocol (MCP) server for AWS S3, enabling AI models, particularly Large Language Models (LLMs), to securely interact with S3 buckets.
This MCP server has critical gaps in definition quality. Only 2 tools are registered, both with empty input schemas (no parameters). While descriptions exist, they are moderately detailed but lack critical parameter-level documentation. The server provides no parameter descriptions, no input validation guidance, no output schema documentation, and minimal error handling. Tool compositions are single-purpose but lack clear chains. No tool annotations present. This is typical of early-stage community tools; it scores in the F-D range.
List all S3 buckets in your storage account (supports AWS S3, DigitalOcean Spaces, IBM Cloud Object Storage, and other S3-compatible services)
Test MCP S3 server connection
No input schemas: Both tools declare empty input objects with no properties. Per hard scoring rules, schema score MUST be 0 when no schema is present.
test_connection tool name is ambiguous. 'test_connection' could mean test a specific connection, or test generic connectivity. Better: 'verify_s3_credentials' or 'check_s3_access', these verbs clearly convey the action and outcome.
No parameter descriptions. While input schemas are empty (no configurable parameters), the tool implementations reference S3 configuration from environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, S3_ENDPOINT_URL, AWS_DEFAULT_REGION). These should be documented in the server's README or tool descriptions to guide users on required setup.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 34 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 31 | - | v1 |
No output schema documentation. list_buckets_tool returns TextContent with unstructured text. LLMs cannot parse or extract structured data (bucket names, creation dates, counts) from plain text. Return structured JSON with fields: buckets (array of {name, created_at}), total_count, service_name. This enables downstream tool chaining and accurate parsing.
Descriptions lack LLM-optimized length and clarity. test_connection is 34 characters ('Test MCP S3 server connection'), below rubric baseline (194 chars average for A+ tools). list_s3_buckets is longer but includes example endpoints (nyc3.digitaloceanspaces.com, s3.us-south.cloud-object-storage.appdomain.cloud), LLMs may reuse these literally instead of adapting to user's actual endpoint. Remove examples; use parameter constraints instead.
Error responses are user-friendly but not LLM-actionable. Example: 'S3 credentials not configured!' followed by setup instructions. This is good UX but does not guide the LLM on what recovery steps to take programmatically. Include recovery hints: 'Call no tools until credentials are configured in the client config.' or 'Agent cannot proceed; ask the user to configure credentials.'
No tool annotations. Neither tool declares readOnlyHint, destructiveHint, or idempotentHint. list_s3_buckets is read-only and idempotent, annotate with {readOnlyHint: true, idempotentHint: true}. test_connection is read-only, annotate with {readOnlyHint: true}. Annotations help agents reason about safety and retry logic.
Incomplete tool composition. list_s3_buckets returns unstructured text with bucket names and creation dates, but no tool exists to interact with buckets (list objects, download file, etc.). From the README references ('download_files'), the server is intended to support file operations but they are not visible in the source. Scope is unclear. Either implement missing tools or update the tool list in documentation.