#authorize anthropic

1 messages · Page 1 of 1 (latest)

short sluiceBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

wary cragBOT
#

For Anthropic subscription auth in OpenClaw (non-API-key), use the setup-token flow:

openclaw models auth setup-token --provider anthropic

If you want the full interactive setup path instead:

openclaw onboard --auth-choice setup-token

You’ll still generate the token via Claude CLI (claude setup-token) and paste it when prompted.

Refs: https://docs.openclaw.ai/providers/anthropic · https://docs.openclaw.ai/concepts/oauth

#

Yep — generate it with the Claude Code CLI:

claude setup-token

That prints a token string. Copy it, then paste it into the openclaw models auth setup-token --provider anthropic prompt.

You can run claude setup-token on any machine. If it’s a different machine than your gateway host, just copy the token over and paste it there.

Ref: https://docs.openclaw.ai/help/faq#where-do-i-find-an-anthropic-setup-token

#

Nice — if you installed Claude Code via npm, it’s usually:

npm uninstall -g @anthropic-ai/claude-code

If that says package not found, check what installed claude first:

which claude
npm ls -g --depth=0 | grep -i claude
pnpm ls -g --depth=0 | grep -i claude

Then uninstall with the same package manager you used (npm uninstall -g <pkg> or pnpm remove -g <pkg>).

#

Perfect — since you used the native installer (curl ... install.sh | bash), here’s the full cleanup.

Uninstall Claude Code (native install: macOS/Linux/WSL)

rm -f ~/.local/bin/claude
rm -rf ~/.local/share/claude

Optional: remove Claude Code config/history too (full wipe)

# user-level settings/state
rm -rf ~/.claude
rm -f ~/.claude.json
# 
# project-level settings (run inside any project where you used Claude Code)
rm -rf .claude
rm -f .mcp.json

If you ever had an old npm install too, remove that as well

npm uninstall -g @anthropic-ai/claude-code

Verify uninstall

which -a claude
#

If uninstall worked, which should return nothing (or only unrelated paths).

For verifying Anthropic setup-token is working in OpenClaw + Discord:

  1. On the gateway host (CLI):
openclaw models status

You should see Anthropic auth/profile available.

  1. In Discord thread:
  • Send: /model status
    This shows active model/provider + auth profile being used.
  • Then send a normal test prompt in that thread; if auth is broken you’ll get an auth error quickly.
#

Yep — easiest path is to pin auth order to the good token profile, so OpenClaw stops trying the old one.

openclaw models auth order set --provider anthropic anthropic:default

Then verify:

openclaw models auth order get --provider anthropic
openclaw models status --probe-provider anthropic

In Discord, run:

# 
/model status

You should see it using the Anthropic profile you pinned.

If you want to fully remove the stale one (anthropic:0auth-01), there isn’t a dedicated models auth remove command yet — delete that profile entry from:

~/.openclaw/agents/main/agent/auth-profiles.json

(make a backup first), then:

openclaw gateway restart
#

Got it — let’s do a tight triage.

Most likely causes here are (a) bad session state, or (b) the thread still trying the stale Anthropic profile.

1) Quick in-Discord reset test

  1. Run /new
  2. Run /model status
  3. Send a tiny test message (ping)

If /model status still shows the old profile, force the good one for this session.

2) On your gateway host, run these and paste output

openclaw --version
# 
openclaw status --all
openclaw models status --probe-provider anthropic

3) Capture the exact runtime error

In one terminal:

openclaw logs --follow

Then send one message in Discord to reproduce, and paste the error lines from logs.

That generic “⚠️ Something went wrong…” is just the surface message — logs will show the real failure (auth/profile/session/tool error).

Refs: https://docs.openclaw.ai/help/faq#first-60-seconds-if-something-is-broken · https://docs.openclaw.ai/cli/models

#

Nice catch — your logs show the root cause exactly:

claude-opus-4.6 ❌ (invalid)
claude-opus-4-6 ✅ (correct)

So /new is failing when OpenClaw hits that bad model ref.

Run this on the gateway host:

openclaw models set anthropic/claude-sonnet-4-6
openclaw models fallbacks list
openclaw models fallbacks remove anthropic/claude-opus-4.6
openclaw models fallbacks add anthropic/claude-opus-4-6
openclaw models status
true temple
#

what specifically am i looking for when i run the commands you send me? i don't want to paste all of my logs since this is a public forum. tell me specifically what to look for.

i'm still getting the same error when running "/new" in the channel.

wary cragBOT
#

100% fair — don’t paste full logs. Here’s exactly what to check.

What to look for (pass/fail)

  1. Any model id with a dot version is wrong
    Bad: claude-opus-4.6, claude-sonnet-4.6
    Good: claude-opus-4-6, claude-sonnet-4-6

  2. Fallbacks must not include dotted model ids

  3. Aliases (like opus, sonnet) must point to dashed ids

Minimal commands + expected result

# 
openclaw models status
  • Look for model lines; they should use -4-6, not .4.6.
openclaw models fallbacks list
  • Should not contain 4.6 with a dot.
openclaw models aliases list
  • Any alias mapping to .4.6 is wrong.
#

If you find dotted refs, fix them

openclaw models set anthropic/claude-sonnet-4-6
openclaw models fallbacks remove anthropic/claude-opus-4.6
openclaw models fallbacks add anthropic/claude-opus-4-6
openclaw models aliases add opus anthropic/claude-opus-4-6
openclaw models aliases add sonnet anthropic/claude-sonnet-4-6

Then retry /new.

Safe log snippet (only one line type)

If it still fails, only look for lines containing:

  • model_not_found
#
  • not_found_error
  • No credentials found

You can share just those single lines (no full log dump).

true temple
#

there are no dots.

wary cragBOT
#

Good signal — if there are truly no dotted model refs now, next likely cause is stale runtime state.

Do this exact sequence:

openclaw --version
openclaw gateway restart
openclaw models status --probe-provider anthropic

Then in Discord:

  1. /new
  2. /model status
  3. send ping

If it still errors, capture only matching error lines (not full logs):

# 
openclaw logs --follow | rg -i "model_not_found|not_found_error|no credentials|session file path|embedded_run_agent_end"

Paste just those matched lines.
That’ll tell us if this is still model resolution vs session-state corruption.

true temple
#

give me a command that doesn't use "rg"

wary cragBOT
#

Yep — use grep instead:

openclaw logs --follow | grep -Ei "model_not_found|not_found_error|no credentials|session file path|embedded_run_agent_end"
#

Yep — now we need to find the exact file still holding 4.6.

Run this (no rg):

grep -RIn -E "claude-(sonnet|opus)-4\.6" ~/.openclaw/openclaw.json ~/.openclaw/agents/main 2>/dev/null

That command should print file paths + line numbers.
Those are the stale entries the runtime is still reading.

Fix each file it reports

sed -i.bak \
-e 's/claude-sonnet-4\.6/claude-sonnet-4-6/g' \
-e 's/claude-opus-4\.6/claude-opus-4-6/g' \
# 
<FILE_FROM_GREP>

Then restart:

openclaw gateway restart

If grep prints nothing but errors still mention 4.6, then the gateway is reading a different state/agent path — run:

openclaw health --verbose

and check which config/state path it reports.

true temple
#

the grep command prints way too much output. hard to decipher what's happening there. simplify for me.

wary cragBOT
#

Totally — use this simpler version that prints only file names (no noisy lines):

grep -RIl --include='*.json' --include='*.json5' -E "claude-(sonnet|opus)-4\.6" ~/.openclaw/openclaw.json ~/.openclaw/agents/main 2>/dev/null

If that returns files, auto-fix all of them with:

grep -RIl --include='*.json' --include='*.json5' -E "claude-(sonnet|opus)-4\.6" ~/.openclaw/openclaw.json ~/.openclaw/agents/main 2>/dev/null \
| while IFS= read -r f; do
sed -i.bak -e 's/claude-sonnet-4\.6/claude-sonnet-4-6/g' -e 's/claude-opus-4\.6/claude-opus-4-6/g' "$f"
echo "fixed: $f"
done

Then restart:

# 
openclaw gateway restart
#

Perfect — that means the stale model ref is in your main config.

Run exactly this:

sed -i.bak \
-e 's/claude-sonnet-4\.6/claude-sonnet-4-6/g' \
-e 's/claude-opus-4\.6/claude-opus-4-6/g' \
/home/claw/.openclaw/openclaw.json

Verify it’s clean:

grep -nE "claude-(sonnet|opus)-4\.6" /home/claw/.openclaw/openclaw.json || echo "✅ no dotted refs left"