Third-party apps now draw from your extra usage, not your plan limits. Add more at claude.ai/settings/usage and keep going.

[ajaypilaniyawynk]

Using this bridge I was able to successfully use claude without any issues since yesterday. But today I am getting below error :

Third-party apps now draw from your extra usage, not your plan limits. Add more at claude.ai/settings/usage and keep going.

Maybe Claude has put some more server side validation.

[dotCipher]

@ajaypilaniyawynk I tested this locally on a Claude Max plan and could not reproduce the extra-usage classification after refreshing OAuth credentials.

What I verified:

• ran the repo's side-by-side validator against official Claude Code and OpenCode + the bridge
• both clients succeeded
• extra_usage stayed disabled before and after both requests
• seven_day_oauth_apps stayed null
• so on this machine, with a fresh login, the bridge was still classified onto normal plan usage rather than extra usage

I did hit stale OAuth credentials before re-logging in, so the first thing I'd suggest is:

1. run claude login again
2. retry the bridge immediately after
3. check whether official claude CLI on the same account also shows the same usage warning

Because I can't repro it now, I don't yet have evidence of a deterministic bridge bug. It may be account/session-specific or a server-side classification change.

If you can still reproduce it after a fresh claude login, the most useful next step would be to run the validator and attach the artifacts:

npm run build
node scripts/validate-opencode-oauth.js --model claude-sonnet-4-6 --prompt "Reply with exactly VALIDATE."

Please attach:

• tmp/.../report.json
• the redacted request/response capture files from that run

That will let us compare official Claude Code vs OpenCode on an affected account and see whether there's still a bridge-controlled request difference worth fixing.

[ajaypilaniyawynk]

This is working now. Another issue I am facing is in plan mode where I get error - Model tried to call unavailable tool 'EnterPlanMode'

@dotCipher is it a known limitation?

[dotCipher]

@ajaypilaniyawynk glad the original issue resolved! The plan mode error is a known bug I just fixed — the bridge was injecting EnterPlanMode and ExitPlanMode into the Claude tool schema but had no translation entry mapping those names back to OpenCode's native plan_enter/plan_exit names. So when the model called EnterPlanMode, OpenCode rejected it with "Model tried to call unavailable tool" because it only knows the tool by plan_enter.

Released in v1.10.8 — update your local install and restart OpenCode:

# clear the opencode plugin cache so it picks up the new version
rm -rf ~/.cache/opencode/node_modules/opencode-claude-bridge
rm -rf ~/.cache/opencode/packages/opencode-claude-bridge@latest

Then restart OpenCode and it will fetch 1.10.8 automatically. Plan mode should now work end-to-end.

[dotCipher]

@ajaypilaniyawynk let me know if that worked for you?

[olosegres]

If someone still has problems, probably you need to remove opencode-anthropic-auth plugin from all opencode.json (user level and project level).

[dotCipher]

@ajaypilaniyawynk going to close this within the next 48 hours if you are unable to confirm that this is still an issue

[joyalkenus]

Thanks — you're right, I can't reproduce it deterministically either.

validator did flag bridge-controlled request diffs vs official Claude Code that may still be worth a look as classification signals: tool count 10 vs 54 (toolNamesEqual=false), user-agent (external, cli) vs (external, sdk-cli), missing beta flags thinking-token-count-2026-05-13 and extended-cache-ttl-2025-04-11, an extra x-session-id header, and stainless package version 0.81.0 vs 0.94.0.

[dotCipher]

Thanks, this was useful. I re-ran the validator after refreshing Claude OAuth and could reproduce the bridge-controlled header diffs you called out.

I opened a PR for the safe request-parity fixes here: #23

What changed in that PR:

• added the current Claude Code beta flags thinking-token-count-2026-05-13 and extended-cache-ttl-2025-04-11
• updated x-stainless-package-version from 0.81.0 to 0.94.0
• stripped OpenCode's extra x-session-id header before forwarding to Anthropic
• improved the validator report so it lists missing/extra beta flags and tool names directly
• added newly observed Claude-only stub schemas for PushNotification and ScheduleWakeup

After the patch, the validator reports:

Header differences: 0
Missing beta flags: 0
Extra beta flags: 0

One important note: on my account, both official Claude Code and OpenCode consumed extra_usage during the validator run, so this result does not show a bridge-only classification difference anymore. The header diffs were real, though, and are now addressed in the PR.

The remaining diff is tool shape: official Claude Code currently advertises 30 tools, while OpenCode advertises its active executable tool set. I am not going to blindly mirror all 30 yet, because advertising tools OpenCode cannot execute can cause model/tool-call failures. That should be handled separately with deliberate translation/shim support where possible.