feat(everything): add tool-level annotations to get-annotated-message

[olaservo]

Description

Add ToolAnnotations to all 18 tools in the Everything example server. Previously only get-annotated-message had annotations; now every tool explicitly declares readOnlyHint, destructiveHint, idempotentHint, and openWorldHint so the server serves as a complete reference for developers implementing tool annotations per the MCP spec.

Server Details

• Server: everything
• Changes to: tools, docs

Motivation and Context

Tool annotations are part of the MCP specification and provide hints to clients about tool behavior:

Annotation	Purpose
readOnlyHint	Indicates the tool does not modify its environment
destructiveHint	Indicates whether the tool may perform destructive updates
idempotentHint	Indicates whether repeated calls with same args have no additional effect
openWorldHint	Indicates whether the tool interacts with external entities

The Everything server is the canonical MCP reference/demo server, yet only 1 of its 18 tools had ToolAnnotations. Adding annotations to all tools makes it a complete reference for:

• Client developers testing annotation support
• Server developers learning proper tool annotation usage
• Integration testing of annotation handling

Tool Annotations Added

Tool	readOnlyHint	destructiveHint	idempotentHint	openWorldHint	Rationale
echo	true	false	true	false	Pure function, echoes input back
get-annotated-message	true	false	true	false	(already existed) Returns annotated messages
get-env	true	false	true	false	Reads process.env, no modifications
get-resource-links	true	false	true	false	Returns links to internal resources
get-resource-reference	true	false	true	false	Returns internal resource references
get-roots-list	true	false	true	false	Reads client roots, no modifications
get-structured-content	true	false	true	false	Returns simulated weather data
get-sum	true	false	true	false	Pure arithmetic computation
get-tiny-image	true	false	true	false	Returns a static encoded image
trigger-long-running-operation	true	false	true	false	Simulates work with progress, no lasting state change
toggle-simulated-logging	false	false	false	false	Toggles server-side logging state (not idempotent)
toggle-subscriber-updates	false	false	false	false	Toggles resource subscription updates (not idempotent)
trigger-elicitation-request	false	false	false	false	Triggers user-facing dialog via client
trigger-elicitation-request-async	false	false	false	false	Async task-based elicitation via client
trigger-sampling-request	false	false	false	true	Triggers LLM inference (external API)
trigger-sampling-request-async	false	false	false	true	Async task-based LLM inference (external API)
gzip-file-as-resource	false	false	true	true	Fetches URLs and creates session resources; same input produces same output
simulate-research-query	false	false	false	false	Creates background tasks with internal state

How Has This Been Tested?

• TypeScript compilation passes (npm run build)
• Verified annotations are accepted by SDK types with no type errors
• Tested with MCP Inspector to confirm annotations appear in tools/list responses

Tested with this version of Inspector in this open PR: modelcontextprotocol/inspector#1066

Breaking Changes

None. Annotations are optional metadata hints and do not affect tool behavior.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Checklist

• I have read the MCP Protocol Documentation
• My changes follows MCP security best practices
• I have updated the documentation accordingly
• I have tested this with an LLM client
• My code follows the repository's style guidelines
• New and existing tests pass locally
• I have added appropriate error handling
• I have documented all environment variables and configuration options

Additional context

The annotation values follow the MCP spec semantics:

• readOnlyHint: true when the tool only returns data with no side effects
• destructiveHint: false for all tools since none delete or overwrite external data
• idempotentHint: true when repeated calls with the same args produce the same result and no additional environmental effect; false for toggles, user-interactive tools, and task-creating tools
• openWorldHint: true only for tools that interact with external systems (URL fetching, LLM sampling); false for tools operating within the closed MCP server/client domain

[olaservo]

Unrelated updates - these were just missing from the docs.