[Bug]: max allowed tokens reached

[eviltik]

What happened?

● spec-workflow - approvals (MCP)(action: "status", approvalId: "approval_1758751983781_vnrvfu3z1")
⎿ Error: MCP tool "approvals" response (26638 tokens) exceeds maximum allowed tokens (25000). Please use pagination, filtering, or limit parameters to reduce the response size.

Component

MCP Server

Steps to Reproduce

I think the error message is enought to understand the problem ;)

Expected Behavior

I don't know. If resume we are loosing infos. We can not remove text (except manualy ?) in the spec so .. i do no.

Error Output (if any)

● spec-workflow - approvals (MCP)(action: "status", approvalId: "approval_1758751983781_vnrvfu3z1")
  ⎿  Error: MCP tool "approvals" response (26638 tokens) exceeds maximum allowed tokens (25000). Please use pagination, 
     filtering, or limit parameters to reduce the response size.

Version

v1.0.1

Operating System

Linux

Additional Context

No response

[Pimzino]

Oh Damn! Mind sharing the approval record? How many annotations did you add lol?

[eviltik]

:)

Let's say 20 specific annotations (short sentences, 4/5 words), 1 short general annotation, and one general annotation, perhaps 70 words)

[Pimzino]

Okay, thats odd that it would be maximizing the approval record. That makes no sense.

Are you able to confirm the the document content is in the approval record?

[Pimzino]

You can configure claude code with an env variable to allow more!

export MAX_MCP_OUTPUT_TOKENS=50000

[eviltik]

Nope. But i just lost everything because claude compact the conversation. I am lost in limbo.

So now i've got this in claude code after compact

═══════════════════════════════════════ Conversation compacted · ctrl+o for history ═══════════════════════════════════════
⎿ Referenced file .spec-workflow/specs/filestore/requirements.md
⎿ Read .spec-workflow/steering/structure.md (205 lines)
⎿ Read .spec-workflow/steering/tech.md (126 lines)
⎿ Read .spec-workflow/steering/product.md (76 lines)
⎿ Read .spec-workflow/templates/structure-template.md (146 lines)
⎿ Todo list read (9 items)

● spec-workflow - approvals (MCP)(action: "status", approvalId: "approval_1758754229267_loj6526yu")
⎿ {
"success": true,
"message": "BLOCKED: Status is pending. Verbal approval is NOT accepted. Use dashboard or VS Code extension only
… +25 lines (ctrl+o to expand)

BUT FINALY !!!!!!!!!

Everything is OK. Let me explain

1. I add annotations then ask for a review
2. error: response (26638 tokens) exceeds maximum allowed tokens (25000).
3. i asked for a resume, but a light resume. Claude give me a resume of 3000 tokens lol, so i "loose" everything.
4. then i make another prompt saying that i want 25000 tokens, not 3000
5. after claude refactor the spec, it compacted the conversation => i think everything will going crazy but not
6. Now, i've got a nice requirement to fix and

It's magic.

So. Perhaps should be great to have a sort of token counter in the bottom of the page and some warnings related to size limit ?

[Pimzino]

Nope. But i just lost everything because claude compact the conversation. I am lost in limbo.

So now i've got this in claude code after compact

═══════════════════════════════════════ Conversation compacted · ctrl+o for history ═══════════════════════════════════════ ⎿ Referenced file .spec-workflow/specs/filestore/requirements.md ⎿ Read .spec-workflow/steering/structure.md (205 lines) ⎿ Read .spec-workflow/steering/tech.md (126 lines) ⎿ Read .spec-workflow/steering/product.md (76 lines) ⎿ Read .spec-workflow/templates/structure-template.md (146 lines) ⎿ Todo list read (9 items)

● spec-workflow - approvals (MCP)(action: "status", approvalId: "approval_1758754229267_loj6526yu") ⎿ { "success": true, "message": "BLOCKED: Status is pending. Verbal approval is NOT accepted. Use dashboard or VS Code extension only … +25 lines (ctrl+o to expand)

BUT FINALY !!!!!!!!!

Everything is OK. Let me explain

1. I add annotations then ask for a review
2. error: response (26638 tokens) exceeds maximum allowed tokens (25000).
3. i asked for a resume, but a light resume. Claude give me a resume of 3000 tokens lol, so i "loose" everything.
4. then i make another prompt saying that i want 25000 tokens, not 3000
5. after claude refactor the spec, it compacted the conversation => i think everything will going crazy but not
6. Now, i've got a nice requirement to fix and

It's magic.

So. Perhaps should be great to have a sort of token counter in the bottom of the page and some warnings related to size limit ?

See my previous response please. This is client related and each MCP client will have its own limits.

[eviltik]

ho, i missed your tips, thank you ! It should be great that this limit appear somewhere in the UI ;)

[Pimzino]

ho, i missed your tips, thank you ! It should be great that this limit appear somewhere in the UI ;)

I will try to add a token counter in the UI. The thing is, this wont be a problem for all, its just for some, I think maybe CC does it to avoid context bloat but they allow the user to change it.

[eviltik]

I'm testing spec-worklow on a nice codebase (1 million lines, front/back 12 dockers, mono repo). I just ask for a new feature and i'm follwing your guides (first time). I am excited about the potential outcome. Sound like a game changer. But for now, i'm not coding or follow coding task, spec time only.

Thank you for this crazy and nice project.

[Pimzino]

I am confident you will get amazing results! This is why I have spent the time I have to create this project and also why I have the amount of downloads and stars we have!! I am so glad you found us and I hope you continue to see the benefits!

Good luck my friend and thank you for raising the issue nonethless!