Mastra AI SDK v5: MastraClient 'Failed to parse stream string' and Cloud ENOENT build errors — how to fix?

Question

Upgraded Mastra to the AI SDK v5 beta: MastraClient throws 'Failed to parse stream string. Invalid code data.' and Mastra Cloud builds fail with ENOENT for @mastra/core's libsql module. What are the fixes?

Originally asked on GitHub: https://github.com/mastra-ai/mastra/issues/5470

AgentArmor team · Accepted Answer

Status: the Mastra Cloud build issue was fixed by the maintainers on 2025-07-15, and MastraClient still only works in AI SDK v4 compatibility mode for now, with frontend v5 support still in progress.

For the MastraClient error `Failed to parse stream string. Invalid code data.`, a maintainer explained that a header is now required so the backend knows the client wants AI SDK v4 compatibility:

```ts
const mastraClient = new MastraClient({
  baseUrl: "http://localhost:4111",
  headers: {
    "x-ai-sdk-compat": "v4"
  }
});
```

One commenter confirmed the earlier beta `0.0.0-ai-v5-20250626003446` worked with MastraClient even without this header, so the requirement arrived with a later change. Another commenter hit the parse error after upgrading and reported that rolling back to the previous release fixed it.

For the Mastra Cloud deployment failure (`ENOENT` when loading `@mastra/core/dist/storage/libsql/index.js` from a pnpm path), the reporter's template project was built from the ai-v5 dependency tag. The maintainers said Cloud was changed so ai-v5 versions deploy properly, and the issue was closed as completed.

General advice: with pre-release beta tags, pin the exact version in `package.json` so client and server use matching `@mastra/core` builds, and re-test after each beta release — the compatibility header requirement and stream parsing behavior changed between betas.

AgentArmor team · Answer

**What @DanielSLew tried on GitHub** (Outcome unclear)

Maintainer stated the backend now requires a header for v4 compatibility and suggested passing `x-ai-sdk-compat: "v4"` when creating the MastraClient. No commenter explicitly confirmed this resolved the parse error.

Source: https://github.com/mastra-ai/mastra/issues/5470#issuecomment-3079691200

AgentArmor team · Answer

**What @izziaraffaele tried on GitHub** (Worked)

After upgrading to the ai-v5 beta, calling agent.stream from the MastraClient against a standalone server produced 'Failed to parse stream string. Invalid code data.'; rolling back to the previous release made it work.

Source: https://github.com/mastra-ai/mastra/issues/5470#issuecomment-3062760259

AgentArmor team · Answer

**What @izziaraffaele tried on GitHub** (Worked)

Reported that beta build 0.0.0-ai-v5-20250626003446 works with MastraClient without the v4 compatibility header.

Source: https://github.com/mastra-ai/mastra/issues/5470#issuecomment-3083032409

AgentArmor team · Answer

**What @FlatMapIO tried on GitHub** (Didn't work)

Upgraded a template project's dependency to the ai-v5 tag and deployed to Mastra Cloud; the build failed with ENOENT when loading @mastra/core's libsql storage module from a pnpm path.

Source: https://github.com/mastra-ai/mastra/issues/5470#issuecomment-3062273131

Mastra AI SDK v5: MastraClient 'Failed to parse stream string' and Cloud ENOENT build errors — how to fix?

5 answers

Add your answer or comment

Mastra AI SDK v5: MastraClient 'Failed to parse stream string' and Cloud ENOENT build errors — how to fix?

5 answers

Add your answer or comment

Related questions