Add GenAI semconv attributes to native OTel spans

[etserend]

The existing otel_middleware and client-side spans only emitted mcp.method.name and jsonrpc.request.id. This enriches them with attributes defined in the OTel GenAI MCP semantic conventions:

• gen_ai.operation.name — operation type derived from the method (execute_tool for tools/call, list_tools for tools/list, read_resource for resources/read, get_prompt for prompts/get, etc.)
• gen_ai.tool.name — tool or prompt name from params.name when present
• mcp.resource.uri — resource URI from params.uri or params.ref.uri
• rpc.system = "mcp" — standard RPC system attribute

Motivation and Context

The SDK already ships otel_middleware in src/mcp/server/runner.py producing SpanKind.SERVER spans with W3C trace context propagation. Aligning the native span attributes with the GenAI semconv spec means every MCP-based framework and every distribution gets conformant telemetry without independently re-implementing span plumbing.

How Has This Been Tested?

Adds three tests to tests/shared/test_otel.py:

• test_client_and_server_spans — tools/call: asserts gen_ai.operation.name=execute_tool, gen_ai.tool.name, rpc.system, and trace context propagation (both spans share the same trace_id)
• test_list_tools_spans — tools/list: asserts gen_ai.operation.name=list_tools and absence of gen_ai.tool.name (no specific tool targeted on a list call)
• test_resource_read_spans — resources/read: asserts gen_ai.operation.name=read_resource and mcp.resource.uri

Types of changes

• New feature (non-breaking change which adds functionality)

Checklist

• I have read the MCP Documentation
• My code follows the repository's style guidelines
• New and existing tests pass locally
• I have added appropriate error handling
• I have added or updated documentation as needed

Additional context

[etserend]

Here is example output

 tools/call
├── Span: 'MCP handle tools/call greet' kind=SERVER
│ trace_id: 0xbd62f33a2a8bca7fce981401e9e1fa37
│ ├── gen_ai.operation.name: 'execute_tool'
│ ├── gen_ai.tool.name: 'greet'
│ ├── jsonrpc.request.id: '1'
│ ├── mcp.method.name: 'tools/call'
│ └── rpc.system: 'mcp'
└── Span: 'MCP send tools/call greet' kind=CLIENT
trace_id: 0xbd62f33a2a8bca7fce981401e9e1fa37 ← same trace, context propagated
├── gen_ai.operation.name: 'execute_tool'
├── gen_ai.tool.name: 'greet'
├── jsonrpc.request.id: '1'
├── mcp.method.name: 'tools/call'
└── rpc.system: 'mcp'

tools/list
├── Span: 'MCP handle tools/list' kind=SERVER
│ trace_id: 0xb34caa9e9dbe2748b4b3e37847bd4545
│ ├── gen_ai.operation.name: 'list_tools'
│ ├── jsonrpc.request.id: '1'
│ ├── mcp.method.name: 'tools/list'
│ └── rpc.system: 'mcp'
└── Span: 'MCP send tools/list' kind=CLIENT
trace_id: 0xb34caa9e9dbe2748b4b3e37847bd4545
├── gen_ai.operation.name: 'list_tools'
├── jsonrpc.request.id: '1'
├── mcp.method.name: 'tools/list'
└── rpc.system: 'mcp'

resources/read
├── Span: 'MCP handle resources/read' kind=SERVER
│ trace_id: 0x6f0d7ecaaea9cb4453f7a9e453f923b6
│ ├── gen_ai.operation.name: 'read_resource'
│ ├── jsonrpc.request.id: '1'
│ ├── mcp.method.name: 'resources/read'
│ ├── mcp.resource.uri: 'test://greeting'
│ └── rpc.system: 'mcp'
└── Span: 'MCP send resources/read' kind=CLIENT
trace_id: 0x6f0d7ecaaea9cb4453f7a9e453f923b6
├── gen_ai.operation.name: 'read_resource'
├── jsonrpc.request.id: '1'
├── mcp.method.name: 'resources/read'
├── mcp.resource.uri: 'test://greeting'
└── rpc.system: 'mcp'

[maxisbey]

Thanks for the PR. This has since landed via #2970. I'm closing this out as part of a backlog cleanup now that v2 is out. Feel free to reopen if this is still relevant.

_{AI Disclaimer}