help
Every error code a first-hour call can return, each with its own meaning/cause/fix.
- bearer_invalid -- The call carried no usable credential: no Authorization: Bearer header, and (for host.tool_publish and friends) no tenant_key argument either.
- tenant_key_missing -- A tenant_key argument was required for this call (no Authorization header was sent on this connection) and none was given, or it wasn't a string.
- tenant_key_invalid -- A tenant_key argument was present but matched no tenant.
- rate_limited -- signup's own per-source rate limit was exceeded.
- signup_paused -- The operator has paused new signups host-wide.
- tool_not_found -- The named tool doesn't exist for this tenant (or, for a shared/cross-tenant call, doesn't exist, isn't shared with you, or your group isn't in its sharing list).
- unknown_kind -- host.tool_publish's kind argument named a kind this host doesn't register.
- spec_too_large -- The spec argument serializes to more bytes than this host allows for one tool.
- invalid_spec -- The kind rejected the spec's shape -- a required field missing, or a field that doesn't parse the way that kind expects.
- invalid_params -- A control-plane call argument was missing, the wrong type, or otherwise malformed -- distinct from invalid_spec, which is about a tool's own spec.
- host_not_allowed -- An http-kind spec (or call) named a URL whose host this host's egress policy refuses to reach.
- handle_taken -- The agent-directory handle you tried to claim is already held by someone else.
- handle_reserved -- The agent-directory handle you tried to claim is reserved by the operator (e.g. admin, host, mcphost, system).
- spec_not_exposed -- A tool was shared with you, but not with expose_spec: true, so host.tool_spec_shared can't read its source.
- service_unavailable -- The host refused the call because a resource it depends on (most commonly disk headroom) is below its safety floor.
- internal -- Something failed on this host's side that isn't one of the named error codes above.
- ask_expired -- The ask's link lapsed (its ttl_s ran out) before the user answered, so host.ask.wait will never return an answer for it.
- ask_answered -- A POST to an ask link was refused 409 because the ask already has a recorded answer; the first answer wins and the body names it.
- choice_invalid -- A POST to an ask link was refused 400 because its choice is not one of the ask's options.
- asks_limit -- host.ask.create was refused because this tenant already has its plan's asks_open_max asks pending (unanswered and unexpired).
- capability_unsupported -- The method belongs to an MCP primitive this host does not implement (resources, prompts, completions, sampling, elicitation, logging or roots), so it answers -32601 as the MCP spec requires for an unadvertised capability.
- notify_kind_unknown -- host.notify.channel_add was refused because kind is not a channel kind this host delivers to. The kinds are: {notify_kinds}.
- agent_not_found -- Check the address with host.agent.lookup, then re-send to a handle that exists.
- alert_body_too_large -- Shorten the alert body to data.limit_bytes or less and re-send.
- already_verified -- This tenant is already verified; nothing was written. The data names the method that verified it.
- anchor_not_found -- Pick one of data.did_you_mean, or call host.help.search for a real anchor.
- archive_corrupt -- Re-run host.export and import the fresh archive; this one is damaged or not an export.
- args_invalid -- Re-send with data.field as data.expected; data.example shows a valid value.
- args_schema_not_inferable -- Declare args_schema explicitly in the spec; the source's argument use cannot be inferred.
- ask_not_found -- List asks with host.ask.list and use an id from data.next_call's result.
- auth_conflict -- Send exactly one credential: either the URL secret or the Authorization header, not two different tenants.
- authorization_required -- Authorize through the OAuth metadata named in the WWW-Authenticate header, then retry with the token.
- bad_response -- The host answered with a body that is not a JSON-RPC message; check the URL points at an mcphost /mcp endpoint.
- banned -- This key is banned; do not retry. Contact support through the support_url on /support.html.
- billing_unavailable -- Billing is not configured on this host; do not retry. Read billing.plans for what the host offers.
- blob_expired -- Revive it with host.blob.touch {id, ttl_s} before the nightly prune, or upload the blob again.
- blob_not_found -- List blobs with host.blob.list and use an id or name that exists for this tenant.
- blob_too_large -- The blob is larger than blob_max_bytes (data.limit); upload a smaller file or move to a plan with a larger blob_max_bytes.
- blob_too_large_inline -- Fetch the blob through the link from host.blob.url instead of reading it inline.
- budget_ceiling_exceeded -- Lower the requested budget value to data.ceiling or below, or move to a plan with a higher ceiling.
- budget_exceeded -- Raise the run budget on data.dimension within the plan ceiling, or reduce the work per call.
- build_failed -- Read data.tail for the build error, fix the requirements in the spec, and republish.
- call_timeout -- Reduce the work per call or raise timeout_s within the plan ceiling, then retry.
- capacity -- Retry after data.retry_after_ms; the host is at its concurrent-call limit.
- channel_closed -- Posting is refused on a closed channel; reads still work. Open a new channel to continue.
- channel_frozen -- Posting is refused on a frozen channel; reads still work. Ask the owner to unfreeze it.
- channel_not_found -- Check data.key against the channels you opened with host.channel.open and use an id or name that exists.
- channel_unavailable -- This call context has no channel backend; call the tool directly instead of from a sandbox.
- chart_invalid_mark -- Re-send with one of data.allowed as the mark.
- claim_email_failed -- The claim email could not be sent; retry later or use the claim link from the response.
- client_not_found -- Check data.client_id against the pending clients list and use one that exists.
- compose_children_exceeded -- Reduce the number of child calls in the compose spec to the plan ceiling.
- compose_depth_exceeded -- Flatten the compose spec; nesting is deeper than the allowed depth.
- compose_fanout_exceeded -- Reduce the for_each fan-out to the allowed number of items.
- compose_fanout_item_failed -- Read data for the failing item's own error and fix or skip that item.
- compose_input_missing -- Pass the input the compose mapping reads, named in the message.
- compose_item_outside_for_each -- Use item only inside a for_each mapping.
- compose_mapping_missing -- Add the missing mapping for the child argument named in the message.
- compose_mapping_type -- Map a value of the child argument's declared type.
- compose_mapping_unresolved -- Fix the mapping path named in data.field so it resolves against the inputs.
- compose_self_call -- A compose tool may not call itself; point the step at a different tool.
- concurrency_exceeded -- Wait for in-flight calls to finish and retry; data.plan names the concurrency ceiling.
- confirm_invalid -- Call the dry run again and repeat with the freshly returned confirm token.
- confirm_mismatch -- Pass the org's exact name as confirm (host.org.members names it); nothing was changed.
- confirm_required -- Call dry_run_first.tool with dry_run_first.args, read would, then repeat with the returned confirm token.
- conflicting_credentials -- Send one credential: drop the tenant_key argument or the Authorization header.
- connection_missing -- Connect the provider named in data.provider with host.connect.start, then create the sync again.
- contact_pending -- A request is already pending or recently denied; wait for the recipient to answer.
- contact_refused -- Send host.agent.contact_request first and wait for acceptance.
- contact_request_not_found -- List pending requests and use an id that exists.
- cursor_invalid -- omit cursor to restart from the first page
- dead_letter_not_found -- List dead letters and use an id that exists for this tenant.
- dependency_advisory -- Pin the package to data.fixed or later and republish.
- dependency_policy -- Remove or replace the requirement named in data.reason; the policy denies it.
- dependency_resolve_failed -- Read data.tail for the resolver error and fix the requirements.
- dns_record_not_found -- Publish the TXT record named in expected, wait for DNS to propagate, then call host.claim.dns.check again.
- docs_mime_unsupported -- Use one of data.allowed mimes; binary files go to host.blob.put.
- docs_not_found -- List documents with host.docs.list and use a name or id that exists.
- docs_provider_error -- The embeddings provider returned an error status; retry later or check the provider config.
- docs_provider_unreachable -- The embeddings provider could not be reached; retry later or check its URL.
- docs_secret_missing -- Set the secret named in data.secret with host.secret_set, then retry.
- docs_too_large -- Shorten the document to data.limit_bytes or split it.
- docs_unavailable -- This call context has no document backend; call the tool directly instead of from a sandbox.
- draft_test_failed -- Read data for the failing draft test and adjust the request.
- draft_unavailable -- This host has no draft model configured; build the spec by hand from host.quickstart.
- drift_review_not_found -- List reviews with host.drift.reviews and use an item id that exists.
- drift_target_not_found -- Declare the table or document first, or use a target that already exists.
- drift_unavailable -- This call context has no drift backend; call the tool directly instead of from a sandbox.
- egress_unavailable -- Network egress is not configured on this host; use a kind or host that does not need it.
- elicit_calls_exceeded -- Ask for fewer elicitations per call; data.limit is the maximum.
- elicit_schema_invalid -- Use only the schema subset in data.allowed and stay within data.limits.
- elicit_timeout -- The client did not answer in time; call again and answer within data.timeout_s.
- elicit_unsupported -- This client or context cannot elicit; read data.reason and ask through host.ask.create instead.
- email_not_owner -- Set target to the owner's verified email address, or add a webhook or slack channel for another destination.
- email_owner_unverified -- Verify the owner email first with host.claim.start; until then mail goes only to the owner's own address.
- email_provider_error -- The email provider refused or never answered after one retry; data.status is its HTTP status. Retry later.
- email_quota_exceeded -- The daily email quota is used; tell your user, and retry after data.resets_at or move to a plan with a higher emails_per_day.
- emulator_unsupported -- Publish with mcphost push and call this host.* verb on the host; mcphost dev does not serve it.
- end_user_assertion_invalid -- Send a fresh end_user_assertion; data.reason names what was wrong.
- end_user_explicit_forbidden -- Drop the explicit end_user subject; this call already carries an end-user identity.
- end_user_not_found -- List end users and use a subject that exists for this tenant.
- end_user_required -- Connect with a verified end-user identity, or pass an explicit end_user subject.
- end_user_revoked -- This end user was revoked; do not retry with that subject.
- env_secret_collision -- Rename the env entry or the secret so no name is used by both.
- event_body_too_large -- Send a smaller event body; the limit is the plan's event_body_bytes_max.
- events_rate_limited -- Slow the delivery rate and retry; the plan caps events_per_minute.
- example_exists -- Rename the tool and propose again.
- example_hidden -- The example failed its nightly proof and is hidden (data.reason, data.red_since); pick another from host.example.list or retry after it is green again.
- example_not_found -- Call host.example.list for the example names and pass one of them as name.
- example_unavailable -- Pass args explicitly; no example could be generated from args_schema.
- exec_unavailable -- The sandbox or exec backend is down; retry later or use a kind that does not need it.
- export_link_expired -- Run host.export again to get a fresh signed link.
- export_not_complete -- Wait for the export run to complete, then import.
- export_too_large -- Reduce the tenant's data below the plan's export_bytes_max, or move to a larger plan.
- feedback_not_found -- Use a feedback id this caller filed.
- field_unknown -- Use a field from data.valid; data.nearest is the closest match.
- fixture_shape_mismatch -- Give the fixture one of the declared outputs named in data.expected.
- fixture_step_ambiguous -- Key the fixture by step number; several steps use that tool.
- fixture_step_unknown -- Key the fixture by a step that exists; data.steps is the step count.
- fixtures_not_applicable -- Fixtures apply to chain tools only; drop them for other kinds.
- fixtures_too_large -- Reduce the fixtures to data.limit_bytes in total.
- forbidden -- This credential may not do that; use a key with the needed access.
- grant_not_found -- List grants and use an id that is still live for this tenant.
- group_not_found -- List groups and use an id that exists.
- handle_not_found -- List handles and use one that exists for this tenant.
- handle_quota_exceeded -- Drop unused handles or move to a plan with a larger handle byte quota.
- handoff_token_expired -- Request a new handoff token and redeem it before it expires.
- handoff_token_invalid -- Request a new handoff token; this one is not valid.
- handoff_token_redeemed -- Request a new handoff token; each token redeems once.
- has_children -- Pass cascade: true to revoke the sub-key's live descendants too (data.count), or revoke them first.
- hook_not_found -- Create the event trigger for this tool first, or use one that exists.
- host_unavailable -- The host did not answer; retry later or check /healthz.
- host_unreachable -- The host could not be reached from this client; check the URL and network, then retry.
- import_fetch_failed -- Check the URL is public and returns CSV or JSON Lines; data.hint names the fix when one is known.
- import_parse_failed -- Fix the source at data.line (RFC 4180 CSV with a header row, or one JSON object per line) and retry.
- import_schema_mismatch -- Rename the source's columns to the table's (data.expected) or add the missing columns; data.skip_reasons has the first failures.
- import_too_large -- Send a smaller file or split it; data.bytes is the byte cap that applied.
- import_url_forbidden -- Use a public https URL; data.host names the host the guard refused and data.reason why.
- index_exists -- Use a different index name or drop the existing index first.
- index_not_found -- Create the index first or use a name from the index list.
- insufficient_scope -- Use a token that carries the scope named in data.scope.
- interrupted -- The server restarted before this call finished; retry the same call (data.restart_id names the restart).
- invalid_example -- Fix the errors listed in data.errors and propose again.
- invalid_token -- Fetch a new token; data.error_description says whether it was expired, malformed or from an unknown issuer.
- invalid_tool_name -- Use a tool name of lowercase letters, digits and underscores within the length limit.
- invalid_webhook_signature -- Re-send with a signature computed over the raw body with the signing secret.
- invite_invalid -- Ask the owner for a new invite link; this one is no longer valid.
- invite_rate_limited -- Try the invite link again later; it reached its hourly join limit.
- invites_quota_exceeded -- Revoke an unused invite or move to a plan with a higher invites_max.
- issuer_already_registered -- Use the existing issuer registration or remove it first.
- issuer_not_found -- List issuers and use a registered one.
- issuer_quota_exceeded -- Remove an unused issuer; data.limit is the plan's maximum.
- issuer_url_invalid -- Use an https:// issuer URL.
- jobs_concurrent -- Wait for a running job to finish; the tenant is at its jobs_concurrent ceiling.
- jobs_concurrent_exceeded -- Wait for a running job to finish; the tenant is at its jobs_concurrent ceiling.
- join_code_invalid -- Ask a client already in the workspace for a fresh join link; host.whoami (data.next_call) prints the live join_url.
- jsonrpc_error -- Read the JSON-RPC error's own code and message returned by the host.
- key_bound_elsewhere -- Use this sub-key only from the session or run it is bound to (data.bound), or delegate a new sub-key with host.key.delegate.
- key_expired -- Create a new sub-key with host.key.mint.
- key_revoked -- Create a new sub-key; this one was revoked.
- kind_mismatch -- Send the kind the spec's shape needs (data.inferred), or change the spec to match data.requested.
- label_taken -- Choose a different sub-key label.
- last_client -- Add another client to the workspace first (host.whoami prints the join_url), then revoke this one.
- leak_test_failed -- A secret got through the self-test: the host's leak block is not working; report data.echo and data.egress to the operator.
- lineage_blocked -- Repeat with confirm: true, or remove the consumers named in data.lineage_blocked first.
- lineage_handle_not_found -- Run the trace again and use the fresh handle.
- lineage_unavailable -- This call context has no lineage backend; call the tool directly instead of from a sandbox.
- listen_streams_max -- Close an open subscriptions/listen stream or move to a plan with a higher listen_streams_max; data.limit is the plan maximum of open listen streams.
- mesh_frozen -- Messaging and posting are paused by the operator for this tenant; do not retry.
- meter_emit_locked -- Another emit-meter run holds the lock; retry when it finishes.
- method_server_to_client -- Do not post this method: the host sends it to the client. Answer it when the host sends it, or read /capabilities.json server_to_client.
- method_unavailable -- This host has not configured that verification method; pick another from /claim/methods.json.
- method_unknown -- Do not post this method: the host does not dispatch it on this protocol revision. Read /capabilities.json for the methods it implements.
- msg_unavailable -- This call context has no messaging backend; call the tool directly instead of from a sandbox.
- namespace_reserved -- Pick another namespace: this one belonged to a tenant that was deleted and is never handed out again.
- namespace_unverified -- Verify the namespace with the registry before publishing.
- no_channel -- Register a channel with host.notify.channel_add, then send the notify again.
- not_a_member -- Call host.org.members for the live member list; this namespace is not one of them.
- not_found -- Use a run_id or url from a completed host.export.
- not_in_org -- Call host.org.create or host.org.join first; this tenant belongs to no org.
- not_needed -- Send the message directly; no contact request is needed.
- not_owner -- Only the org owner can do this; ask the owner listed by host.org.members.
- not_portable -- Remove the reference to your tenant named in data.reference and propose again.
- notify_channel_limit -- Remove a channel with host.notify.channel_remove or move to a plan with a higher notify_channels_max.
- notify_channel_not_found -- List channels with host.notify.channel_list and use a channel_id that exists on this tenant.
- notify_limit -- The daily notify quota is used; retry after data.resets_unix or move to a plan with a higher notifies_per_day.
- notify_target_invalid -- Fix target for the channel kind; a webhook or slack target must be an https URL on a public host.
- org_export_running -- An org export is running; poll host.runs.get with data.run_id until it finishes, then offboard.
- org_member_already -- This tenant is already in an org; call host.org.members to see it.
- org_members_max -- Upgrade to the plan named in data.upgrade, or remove a member first.
- org_name_taken -- Choose a different org name; this one is already taken.
- org_required -- Call host.org.create (or host.org.join) first; org usage needs an org.
- org_tool_name_taken -- Another org member already shares a tool under this name as org.<name>; rename yours or ask data.publisher to unshare theirs.
- owner_cannot_leave -- The owner cannot be removed from its own org; remove other members or delete the org instead.
- owner_unverified -- Verify the owner email first with host.claim.start, or add a webhook or slack channel instead.
- owner_verified_by_other_method -- This tenant's owner already proved themselves by another method; the admin path never overwrites a human's or a domain's proof. Nothing was written.
- permission_denied -- Use a key allowed to call this tool; data.allowed_tools lists what this sub-key may call.
- plan_limit -- Wait for the daily limit to reset or move to a plan with a higher limit.
- plan_required -- Move to a plan that includes this feature; data names it.
- policy_widening -- Repeat with replace: true to confirm the wider rule, or narrow the proposed rule.
- proof_failed -- Fix the tool so proof_call holds; data.actual is the result it returned.
- provider_discovery_failed -- Check the issuer URL serves a discovery document, then retry.
- provider_not_found -- Register an OIDC provider first.
- provider_unavailable -- The host has no OAuth client for this provider yet; the operator sets its MCPHOST_CONNECT_<PROVIDER>_CLIENT_ID and _CLIENT_SECRET (host.connect.providers shows the reason).
- quota_blob_bytes -- Delete blobs or move to a plan with a higher blob_bytes_max; data.limit is the plan's maximum.
- quota_caller -- The caller exceeded its calls_per_day limit for this shared tool; retry after the daily reset.
- quota_docs -- Delete a document or move to a plan with a higher docs_max.
- quota_docs_bytes -- Delete documents or move to a plan with a higher docs_bytes_max.
- quota_end_users -- Revoke unused end users or move to a plan with a higher end_users_max.
- quota_exceeded -- Read data for the exceeded quota; reduce usage or move to a plan with a higher limit.
- quota_index_max -- Drop an unused index or move to a plan with a higher index_max.
- quota_syncs_max -- Delete a sync with host.sync.delete or move to a plan with a higher syncs_max; data.limit and data.used say where you are.
- quota_trusted_issuers -- Remove a trusted issuer; data.limit is the plan's maximum.
- quota_vault_providers -- Remove an unused vault provider or move to a plan with a higher vault_providers_max.
- rate_limited_host -- The host is shedding load; retry shortly.
- recipient_inbox_full -- Try again later; that agent's inbox is full.
- registry_disabled -- This host has no registry enabled; do not retry.
- registry_rejected -- The registry refused the publish; correct the spec and retry.
- remember_too_long -- Shorten remember to 4096 bytes or fewer.
- replay_needs_secret -- Pass the full replay.args; the recorded request was too large to keep.
- request_body_too_large -- Send less than data.limit_bytes per call; store a large payload with host.blob.put and pass its blob id.
- requirement_not_allowed -- Use only requirements allowed by the PyPI policy.
- requirement_not_inferable -- List requirements explicitly in the spec; the import cannot be mapped.
- resources_max -- Unsubscribe from a resource or move to a plan with a higher resources_max; data.limit is the plan maximum of open resource subscriptions.
- response_too_large -- Ask the upstream for a smaller response; the cap is in the message.
- revoke_required -- Revoke the end user first, then purge.
- run_error -- Read the run's error_class and fix the tool.
- run_not_cancellable -- The run is already finished or unknown; nothing to cancel.
- run_not_found -- List runs and use an id that exists for this tenant.
- sampling_calls_exceeded -- Ask the client's model fewer times per run; data.limit is the per-run sampling-call ceiling.
- sampling_quota_exceeded -- Wait for the daily reset or move to a plan with a larger sampling_tokens_per_day.
- sampling_timeout -- Retry with a smaller max_tokens or a shorter prompt; the client's model did not answer in time.
- sampling_unsupported -- Do not retry; this client cannot sample. Use data.alternative or call from a client that supports sampling.
- sandbox_unavailable -- Retry in data.retry_after_s seconds or use http instead; the python kind is unavailable.
- sandbox_unsupported -- No registered kind on this host has a sandbox self-test; do not retry.
- schedule_too_frequent -- Use an every schedule that never fires more often than data.min_seconds.
- scope_escalation -- Request only scopes, tools and limits inside the delegating key's own authority; data.excess names the one over.
- scope_missing -- The share does not grant this scope; data.granted lists what it does, ask the owner to re-share with data.required.
- scope_unknown -- Pass only scope names host.connect.providers lists for this provider, or omit scopes for the default set.
- secret_leak -- Replace the literal value with the secret placeholder named in the message and republish.
- secret_leak_blocked -- Stop the tool from putting the secret in data.where (url, header, body, result, excerpt or log), or bind the secret to the right host with host.secret.set hosts.
- secret_missing -- Set the missing secret with host.secret_set, then retry.
- secret_unavailable -- Set the missing secret with host.secret.set, then retry.
- share_quota_exceeded -- Unshare a tool or move to a plan with a higher share limit.
- share_revoked -- The owner stopped sharing this tool; ask the owner to share it again.
- signature_invalid -- Re-send with a valid signature, token and timestamp for the verify config.
- signature_mismatch -- Re-send with a signature that matches the verify config.
- signup_rate_limited -- Retry after data.retry_after_s; too many tenants were created from this address.
- state_quota -- Free state space or move to a plan with a larger state quota.
- state_quota_exceeded -- Delete state rows or move to a plan with a higher state_bytes_max.
- state_table_not_found -- Declare the table first or use a declared name.
- state_unavailable -- This call context has no state backend; call the tool directly instead of from a sandbox.
- step_tool_not_allowed -- Use a host step from the allowlist named in the message.
- step_tool_not_found -- Use a tool that exists; data.suggestions lists near matches.
- storage -- Retry once; if it persists, report data.request_id through the support link.
- table_already_exists -- Use a different table name or drop the existing one.
- table_bound_exceeded -- Narrow the query so it stays under data.bound's limit.
- table_column_not_found -- Use a declared column; data.candidates lists the closest ones in data.table.
- table_not_found -- Declare the table first or use a declared name.
- table_query_rejected -- Fix the query; data.reason says why it was rejected.
- table_row_conflict -- Change the conflicting value in data.column or query the table first; nothing was appended.
- table_unavailable -- This call context has no table backend; call the tool directly instead of from a sandbox.
- task_already_claimed -- Pick another task; this one is already claimed.
- task_closed -- The task takes no further claim or completion; post a new task.
- task_not_claimed -- Claim the task with host.task.claim before completing it.
- task_not_found -- List tasks and use an id that exists.
- template_error -- Fix the template in the field named in data.field.
- tenant_disabled -- Read data.disabled_reason. self_offboard: this account was closed by self_offboard; call any host.* with no key on a new session (or signup) to start fresh. admin_disable: do not retry.
- tenant_key_required -- Call with the tenant key only; this tool refuses an end-user identity.
- tenant_not_empty -- Import into a new empty tenant; this one already holds data.
- tenant_not_found -- Check the namespace spelling and use a tenant that exists.
- thread_not_found -- List threads and use an id that exists.
- timeout -- The host timed out; retry shortly.
- timeout_clamped -- Lower timeout_s to data.ceiling (the plan's sandbox_cpu_s_max) or upgrade the plan; calls already run with the ceiling.
- too_large -- Shrink the payload to the limit in data.max_bytes.
- too_many_invocations -- Send at most data.limit invocations per call.
- too_many_proposals -- Wait for an open proposal to be decided; data.limit is the maximum.
- too_many_rows -- Narrow since/until so the export holds fewer rows than data.max_rows.
- too_many_tools -- Remove a tool or move to a plan with a higher tool limit.
- tool_building -- Retry shortly; the tool's environment is still building.
- tool_exception -- Read data.exception_class and data.line, fix the tool's code, republish and re-run host.tool_test.
- tool_missing -- Restore the target tool or remove the trigger.
- tool_oom -- Reduce the tool's memory use or the input size.
- tool_output_invalid -- Return JSON that matches the tool's declared output.
- tool_output_too_large -- Return less data; the output exceeds the cap.
- tool_process_limit -- Reduce the number of processes or threads the tool starts.
- tool_run_unsupported -- This kind cannot be run this way; call the tool through host.tool_call.
- tool_timeout -- Make the tool finish within the time cap, or reduce its input.
- tool_trapped -- The tool trapped; fix the runtime fault in its code and republish.
- trace_not_found -- Use a call_id from a recent call; traces expire.
- transaction_nested -- Retrying fails the same way; report data.request_id through the support link.
- trigger_interval_too_short -- Use a schedule at or above the plan's schedule_min_interval_s.
- trigger_invalid -- Fix the trigger field named in data.field.
- trigger_kind_mismatch -- Remove the existing trigger of the other kind before reusing the name.
- trigger_not_found -- List triggers and use an id that exists.
- trigger_quota_exceeded -- Remove a trigger or move to a plan with a higher event_triggers_max.
- typescript_not_supported -- Strip the types and publish plain JavaScript (ES2020); data.line is the first line with TypeScript syntax.
- unknown_argument -- Remove the argument; data.known lists the recognised ones.
- unknown_import -- Use a module from the hint in data.hint; the import is not available in the sandbox.
- unknown_spec_field -- Remove the spec field or use one the kind documents.
- untested_tool_cannot_be_public -- Read data.verdict and data.detail, fix the tool, pass host.tool_test (data.next) for its current version, then set visibility public again.
- unverifiable -- Call data.call_instead for real; no dry run exists for this verb.
- upstream_auth -- The upstream rejected the credential (HTTP 401/403). If the tool sends a secret, set or rotate it with host.secret.set, then retry; otherwise reconnect the upstream.
- upstream_client_error -- Fix the request the tool sends; the upstream rejected it as a client error.
- upstream_error -- Retry later; the upstream failed with a server error.
- upstream_not_connected -- Connect the upstream through the vault link first.
- upstream_rate_limited -- Wait and retry; the upstream is rate limiting this tool.
- upstream_status -- Read data.upstream_status for the upstream's HTTP status, then fix the request or retry later.
- upstream_timeout -- Retry later; the upstream did not answer in time.
- upstream_unreachable -- Retry later; the upstream could not be reached.
- uptime_unavailable -- This host has no http kind registered to probe with; do not retry.
- validation -- Fix the counter or field named in data.key.
- vault_provider_not_found -- Register the vault provider first or use a registered name.
- webhook_body_too_large -- Send a body under data.limit_bytes.
- webhook_not_found -- Use a webhook that exists.
- webhook_over_quota -- Retry after the daily reset; the tenant is over its calls_per_day quota.
- webhook_row_not_found -- Use an inbox row id that exists for this webhook.
- webhook_signature_invalid -- Re-send with a signature that matches the webhook secret.
- wrong_tenant -- Use the credential of the tenant this URL belongs to.