> ## Documentation Index > Fetch the complete documentation index at: https://docs.argil.ai/llms.txt > Use this file to discover all available pages before exploring further. # Troubleshooting > Common problems, error codes, security and technical details. ## Common problems > [accordion] The client cannot connect, or gets 401 > > The API key is missing, mistyped, or revoked. Check that `ARGIL_API_KEY` is set in the shell that starts the client, or create a new key in **Workspace settings → API keys**. A chat app that says "This sign-in request has expired" took too long to sign in: connect again. > [accordion] The assistant uses the wrong workspace > > One connection uses one workspace: an API key's, or the one you picked when signing in. Disconnect and connect again with the other workspace, or use a key created in it. > [accordion] A video seems stuck > > Videos take minutes. Ask "check on my last Argil video": the generation kept running even if the conversation stopped. You also find it in your Argil playground. > [accordion] "Plan required" or "insufficient credits" > > Generating needs an active Argil plan and enough credits in the workspace. Both are managed in the [Argil app](https://www.argil.ai); everything else works without them. > [accordion] The client asks before every check on a video > > Allow the read tools (`get_generation`, `get_asset`, `list_*`) in your client's settings. Keep approval on for `generate_media`, which spends credits. > [accordion] The assistant cannot use a file I attached > > Chat apps do not pass attachments to Argil. Upload the file in your Argil library and say "use my latest upload", or give a public link. ## Error codes A refused tool call returns `isError: true` with the reason, and nothing is created or charged. The assistant reads it and usually fixes the call itself. | `code` | Meaning | What to do | | - | - | - | | `parameter_invalid`, `parameter_missing`, `parameter_unknown` | A field is wrong for that model; the answer gives the allowed values | Fix the field (`get_model` with the mode lists them) | | `model_not_found`, `mode_not_supported` | The model or mode does not exist, or is not enabled in your workspace | Pick another with `search_models` | | `reference_not_found`, `reference_not_allowed` | Unknown asset or element, or a file this mode cannot use | Find the right one with `list_assets` or `list_elements` | | `asset_not_ready`, `asset_not_uploaded`, `url_unreachable` | A file is still processing, was never uploaded, or its URL cannot be downloaded | Wait with `get_asset`, upload it, or use a public link | | `credit_limit_exceeded` | The price is above `max_credits` | Raise the budget, or pick cheaper options | | `insufficient_credits`, `plan_required` | Not enough credits, or no active plan | Manage them in the Argil app | | `idempotency_key_reused`, `idempotency_key_in_use` | The key belongs to another request, or its first call is still running | Use a new key, or retry a few seconds later | | `resource_not_found` | No such id in your workspace | List to find the right id | | `generation_not_terminal`, `too_many_webhook_endpoints`, `payload_too_large` | A running generation cannot be deleted; the workspace has too many webhook endpoints; an inline file is over 2 MB | Wait; delete an endpoint; use a URL or an upload | | `rate_limited`, `too_many_queued_generations`, `too_many_asset_imports` | Too much at once | Wait and retry | | `internal_error`, `service_unavailable` | A problem on Argil's side | Retry after a short wait | A generation that starts and then fails, for example on moderation, is not an error code: it ends with `status: failed` and a reason, and its credits are refunded. ## Security and privacy * **One workspace.** A connection works only on its workspace's models, library, elements, credits and webhook endpoints. * **No account access.** An assistant cannot read or create API keys, change your plan or settings, or buy credits. * **Spending is capped by you.** Only `generate_media` spends credits; `max_credits` caps each generation. * **Keys.** Use one key per agent or machine, keep it in an environment variable or a secret store, never in a prompt. Revoking a key stops it at once; generations already running finish. * **Private networks are out of reach.** A URL you give is downloaded by Argil's servers, only from public addresses. * **What is logged.** Each call's tool, duration, outcome, workspace and key. Not your prompts, inputs or results: they stay in your workspace, as in the app. * **Text from your workspace** (prompts, names, descriptions) is returned as data, never as instructions to the assistant. ## Technical details | | | | - | - | | URL | `https://mcp.argil.ai/mcp` | | Transport | Streamable HTTP, stateless | | Authentication | `Authorization: Bearer ` (or `x-api-key`), or OAuth 2.1 starting at `https://mcp.argil.ai/.well-known/oauth-protected-resource/mcp` | | Longest call | About 40 seconds (`wait_seconds`); longer work returns `processing` | | Rate limits | 60 writes and 600 reads per minute per workspace | | Request size | 4 MB; inline files up to 2 MB, uploads up to 2 GB | | Browsers | Requests from web pages (a browser `Origin`) are refused | This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.