MemPalace/mempalace · error · LLMError

Anthropic provider requires ANTHROPIC_API_KEY env or --llm-a

Error message

Anthropic provider requires ANTHROPIC_API_KEY env or --llm-api-key

What it means

LLMError raised by AnthropicProvider.classify() when no API key is configured. The provider is constructed lazily-valid (check_available deliberately skips probing to avoid paid calls), so the missing key surfaces only at the first real classify() request via the X-API-Key header requirement.

Source

Thrown at mempalace/llm_client.py:410

            api_key_source=source,
        )

    def check_available(self) -> tuple[bool, str]:
        if not self.api_key:
            return False, "ANTHROPIC_API_KEY not set (use --llm-api-key or env)"
        # Don't probe — a live request would cost money. First real call will
        # surface auth errors if the key is invalid.
        return True, "ok"

    def classify(
        self,
        system: str,
        user: str,
        json_mode: bool = True,
        think: Optional[bool] = None,  # noqa: ARG002 — accepted for interface compat; Anthropic extended thinking is configured separately
    ) -> LLMResponse:
        if not self.api_key:
            raise LLMError("Anthropic provider requires ANTHROPIC_API_KEY env or --llm-api-key")
        sys_prompt = system
        if json_mode:
            sys_prompt += "\n\nRespond with valid JSON only, no prose."
        body = {
            "model": self.model,
            "max_tokens": 2048,
            "temperature": 0.1,
            "system": sys_prompt,
            "messages": [{"role": "user", "content": user}],
        }
        headers = {
            "X-API-Key": self.api_key,
            "anthropic-version": self.API_VERSION,
        }
        data = _http_post_json(
            f"{self.endpoint}/v1/messages", body, headers=headers, timeout=self.timeout
        )
        try:

View on GitHub (pinned to 06cb6987f0)

Solutions

Export ANTHROPIC_API_KEY in the environment the process actually runs in
Or pass --llm-api-key / api_key= explicitly when building the provider
For local-only use, switch to the ollama provider instead — no key needed
Add a startup check: provider.api_key is truthy before the first classify

Example fix

# before
provider = build_provider("anthropic", model="claude-sonnet-4-20250514")
provider.classify(s, u)  # LLMError: Anthropic provider requires ANTHROPIC_API_KEY env or --llm-api-key

# after
export ANTHROPIC_API_KEY=sk-ant-...
provider = build_provider("anthropic", model="claude-sonnet-4-20250514")
provider.classify(s, u)

Defensive patterns

Strategy: validation

Validate before calling

import os

if os.environ.get("MP_LLM_PROVIDER") == "anthropic" and not os.environ.get("ANTHROPIC_API_KEY"):
    raise RuntimeError("ANTHROPIC_API_KEY not set — refusing to start BYOK path")

Try / catch

from mempalace.llm_client import LLMError

try:
    provider.classify(s, u)
except LLMError as e:
    if "ANTHROPIC_API_KEY" in str(e):
        raise SystemExit("Set ANTHROPIC_API_KEY or switch to the local ollama provider")
    raise

Prevention

Check provider.api_key is set right after construction for anthropic
Inject secrets via the environment of the actual process (systemd, CI secret store)
Default to local providers (ollama) so no key is ever required

When it happens

Trigger: build_provider("anthropic", model=...) without ANTHROPIC_API_KEY in the environment and without --llm-api-key; the key env var set in a different shell/session than the one running the pipeline.

Common situations: Enabling the external BYOK Anthropic path but forgetting the env var; running under systemd/cron where the interactive shell's exports are absent; CI lacking the secret.

Related errors

AI-assisted analysis of MemPalace/mempalace@06cb6987f0 (2026-08-15). Data as JSON: /api/errors/7e0d292708a33e06. Report an issue: GitHub.