feat(server): migrate claude-provider to @protolabsai/sdk/anthropic-compat

[mabry1985]

Why

PR 3 of the namesake SDK cutover. PRs #3625 + #3626 added ProtoProvider and made it the primary driver. This PR migrates the runtime query() call in ClaudeProvider from @anthropic-ai/claude-agent-sdk to @protolabsai/sdk/anthropic-compat — the drop-in compatibility subpath we shipped in protoCLI #241 + released as @protolabsai/sdk@0.3.0.

Net effect: the only remaining caller of @anthropic-ai/claude-agent-sdk is sdk-options.ts, which still uses it for type imports only. PR 4 finishes that and removes the dep entirely.

Changes

apps/server/package.json — bump @protolabsai/sdk to ^0.3.0 (the version with the anthropic-compat subpath).

apps/server/src/providers/claude-provider.ts — swap the imports:

- import { query, type Options, ... } from '@anthropic-ai/claude-agent-sdk';
+ import { query, type Options, ... } from '@protolabsai/sdk/anthropic-compat';

Three small adapter fixes for shape mismatches between our ExecuteOptions and the compat layer's narrower Options:

• systemPrompt — only passed through when it's a string. The SystemPromptPreset shape (Claude SDK's { preset: 'claude_code' }) isn't surfaced through the compat layer; callers needing presets should provide a string at the ExecuteOptions layer.
• env — buildEnv() returns Record<string, string | undefined>. The compat layer's env field is Record<string, string>. Filter undefined values before assignment.
• mcpServers + outputFormat — cast to compat's looser types. The runtime translation in the compat layer passes both through verbatim; the type-level cast just bridges the structured McpServerConfig union from @protolabsai/types to the compat layer's Record<string, unknown> shape.

apps/server/tests/unit/providers/claude-provider.test.ts — vi.mock target updated to @protolabsai/sdk/anthropic-compat so the test mocks intercept the same import the production code now uses.

Out of scope (deferred to PR 4)

apps/server/src/lib/sdk-options.ts still imports Options, HookCallback, PostToolUseHookInput from @anthropic-ai/claude-agent-sdk. The compat layer's current typing is too narrow to absorb these as-is:

• HookCallback in the compat layer is (input) => ... (1 arg); Claude SDK's actual signature is (input, toolUseID, options) => ... (3 args).
• mcpServers field types differ between @protolabsai/types's structured McpServerConfig union and the compat's Record<string, unknown>.
• systemPrompt field accepts string | SystemPromptConfig in usage; compat accepts only string.
• outputFormat field accepts a structured { type: 'json_schema', schema }; compat accepts only string.

Each of these is a widening change to @protolabsai/sdk/anthropic-compat's .d.ts.template in protoCLI. Filing follow-ups; until they ship in @protolabsai/sdk@0.3.1, sdk-options.ts keeps its current import path. Functionally fine — the type-level imports compile to no runtime code, and the runtime query() call on the chat/agent path now goes through the compat layer via ClaudeProvider.

Validation

• npm run typecheck clean across all 21 workspaces
• vitest run in apps/server: 3431 passed, 23 skipped, 0 failed
• claude-provider's own 34 tests pass with the new mock target

After this lands

PR 4 (follow-up, after compat layer widening in protoCLI):

1. Widen HookCallback, mcpServers, systemPrompt, outputFormat types in the compat .d.ts.template
2. Release @protolabsai/sdk@0.3.1
3. Migrate sdk-options.ts imports
4. Delete verify-claude-auth.ts + UI call sites
5. Remove @anthropic-ai/claude-agent-sdk from apps/server/package.json

[coderabbitai]

Warning

Review limit reached

@mabry1985, we couldn't start this review because you've used your available PR reviews for now.

Your plan currently allows 2 reviews/hour. Refill in 15 seconds.

Your organization has run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After more review capacity refills, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans have higher rate limits than trial, open-source, and free plans. In all cases, review capacity refills continuously over time.

Please see our FAQ for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 6ef9304b-fb61-4b1d-bc69-8122aecb9326

📥 Commits

Reviewing files that changed from the base of the PR and between 43003f3 and 7e9f0bf.

⛔ Files ignored due to path filters (1)

• package-lock.json is excluded by !**/package-lock.json

📒 Files selected for processing (4)

• apps/server/package.json
• apps/server/src/lib/sdk-options.ts
• apps/server/src/providers/claude-provider.ts
• apps/server/tests/unit/providers/claude-provider.test.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/rip-anthropic-sdk-final-cutover

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

Code Review — ? finding(s)

Async review running parallel to CodeRabbit. Findings are advisory; not all are merge blockers.

protoLabs Code Review Report

• Generated: 2026-05-22T22:34:31Z
• Git head: fb35454b4d1f108d7d6ec95a0277b089c867d954
• Features mapped: 3
• Findings: 0

No findings recorded.