[Shweta-Deshpande]

What

The server now runs on the official MCP Go SDK by default, for both stdio and
streamable HTTP. The old mark3labs server is still there as a fallback.

Why

This is the "Default server to the new SDK implementation" step of the SDK
migration. It has to land before the mark3labs code can be removed.

Changes

• New mcpofficial.Enabled(). The SDK server is used unless
  TF_X_OFFICIAL_SDK_ENABLED is set to false. Unset, empty or misspelled
  values keep the SDK server, so a typo can't silently pick the old one.
• Stdio and HTTP both use it. On HTTP the SDK server answers on /mcp. The old
  /mcp/official path still works as an alias.
• The SDK server reports the name terraform-mcp-server, same as the old one.
• Tools that need a Terraform token are hidden from tools/list when the
  request has no token, same as the old server. Calling one still returns the
  normal "no token" error.
• e2e, hcpt and workflows now test the SDK server by default, with the old
  server as the second matrix entry. Makefile: test-e2e-legacy replaces
  test-e2e-official-public-registry.
• Added an env var row in the README.

Rollback

Set TF_X_OFFICIAL_SDK_ENABLED=false and restart. The server logs a warning
at startup when the fallback is active. The fallback is temporary.

What users may notice

• Tool responses that used a JSON:API envelope are now flat JSON.
• Hand-written HTTP clients must send Accept: application/json, text/event-stream.
• list_stacks with no match returns an empty list, not an error.

PCI review checklist

• I have documented a clear reason for, and description of, the change I am making.

• If applicable, I've documented a plan to revert these changes if they require more than reverting the pull request.

• If applicable, I've documented the impact of any changes to security controls.

  Examples of changes to security controls include using new access control methods, adding or removing logging pipelines, etc.

[Marin2409]

nitpick: The test step still references the removed matrix.endpoint, so both jobs end up connecting to http://localhost:8080 instead of /mcp. It also doesn't receive TF_X_OFFICIAL_SDK_ENABLED, which causes officialEndpoint() to classify the legacy matrix entry as official.

Could we set both environment variables on the Run tests against HCPT step?

[Shweta-Deshpande]

Ah good catch. Done.

[Marin2409]

The CORS test now runs against the official SDK, but its direct HTTP requests are missing the required Accept header. This makes the server return HTTP 400 and is causing CI to fail.

Could we add Accept: application/json, text/event-stream to both the initialization request and the follow-up tool request?

Inside initializeMCPSession, near line 262, add the header after Content-Type:

req.Header.Set("Content-Type", "application/json")
req.Header.Set("Accept", "application/json, text/event-stream")

then, inside testCORSMode, near line 173, add the same header:

req.Header.Set("Content-Type", "application/json")
req.Header.Set("Accept", "application/json, text/event-stream")

Also the outdated comment about using only the legacy endpoint should also be updated to something:
// Run the CORS checks against the server selected for this test run.

[Shweta-Deshpande]

Done. Also added checks to read JSON payload of a response that is either plain JSON (mark3labs) or a single SSE message with a "data:" line (go-sdk)