Troubleshooting

Six problems cover nearly every support thread. Each section says what you're seeing, what it means, and what to do. If none fits, email team@ansvar.eu with the tool call and the exact error text.

The OAuth window never opens

Your client is blocking the popup, or your network blocks auth.ansvar.eu. Allow popups for the client, check outbound HTTPS to auth.ansvar.eu, and retry the connect. CLI clients (Claude Code /mcp, Gemini CLI /mcp auth ansvar) print the URL — open it by hand if the browser doesn't launch.

Works for a few minutes, then 401s

A pasted access token expired — gateway tokens live minutes by design. Don't paste tokens: connect via the client's OAuth flow (tokens then refresh silently), use the mcp-remote bridge, or a managed service credential for headless setups — see Clients without OAuth. If OAuth itself was working and a long session starts failing, your client may hold a stale token: reconnect (Gemini CLI needs v0.44.0+ for mid-session refresh — see Connect Gemini).

404 on connect

The MCP endpoint is https://gateway.ansvar.eu/mcp — the bare host without /mcp returns 404. Also restart the client after editing config; most MCP clients read it only at launch.

"Method not found" (JSON-RPC -32601)

The tool exists but not for your tier — tier-gated tools are absent from tools/list and rejected if called anyway. The classic case: an agent calls start_workflow on a Free or Premium account (workflows are Team+). It's a tier gate, not an outage. Check the tool reference for what each tier includes, or ask your agent to run describe_capabilities.

"cap_exceeded" (JSON-RPC -32000)

A daily quota or concurrency cap. The error is structured — your agent can read it:

{ "code": -32000, "data": { "cause": "cap_exceeded", "...": "..." } }

Daily search quotas: Free 100/day; Premium 5,000 per seat/day; Team 50,000 and Company 500,000 pooled per organization. Quotas reset daily; per-minute rate limits are separate and much higher. Run get_my_capabilities for your account's live numbers.

Zero results — but no error

The gateway fails closed: real failures come back as errors, so a clean empty result means the query genuinely matched nothing. Three usual causes:

• No scope. search requires at least one of jurisdictions, frameworks, sector, or sources.
• Compound query. Full-text search matches ALL terms — "incident reporting deadline financial entities DORA" easily matches nothing. Decompose into shorter queries ("incident reporting", scoped to the DORA framework) and combine results.
• Out of coverage. Check list_coverage(jurisdiction) — a well-behaved agent reports "queried, zero results, not in coverage" rather than answering from memory.

"All resolved MCP endpoints failed"

A downstream corpus was unreachable and the gateway refused to pretend otherwise (accuracy over availability — you get an error, never a silently thinner answer). It's usually transient; retry once. If it persists for a specific jurisdiction, report it — that error names a real outage, not a client problem.