| alerts.channels.list | List the budget alert channels. A webhook target is masked; the URL is a secret. | none |
| audit.list | List audit-log entries, newest first: who changed what, when, and the before and after. | none |
| auth.whoami | Describe the calling principal and the operations it is allowed to call. | none |
| batches.list | List batch jobs, newest first, each with its per-model usage ledger and deferred cost. | none |
| config.generations.list | List the routing-config history, newest first: one entry per change to the document. | none |
| config.generations.read | Read one past routing-config state as a document, ready to validate and apply back. | none |
| config.read | Read the routing config as a declarative, round-trippable document. | none |
| config.utterances.read | Read one route's classifier examples. 404 for an unknown route. | none |
| config.validate | Dry-run a config document: diagnostics plus a structured diff against the live config. | none |
| connections.list | List every upstream connection: whose it is, the provider, scopes, state and last use. Never a token. | none |
| connections.providers.list | List the OAuth providers a person may connect an account at, with their endpoints. Never the client secret. | none |
| credentials.list | List provider credentials: name, provider and endpoint. Never the provider key. | none |
| credentials.usage | Read each provider credential's usage in its limit window, as tokens, USD and percent of the limit. | none |
| guardrails.activity | Count the requests each guardrail rule acted on, per stage and action, over a window (default: 30 days). | none |
| guardrails.read | Read the screening rules: the master switch, every rule with its scope and mode, soft-deny and stream screening. | none |
| keys.list | List client keys with their owner, route and ZDR fences, tool grant, team, budget and spend. Never a secret. | none |
| logs.list | List recent request-log rows, newest first: route, key, model, status, latency, tokens and cost. Never a body. | none |
| model-policy.list | List the per-model zero-data-retention attestations that ZDR-only keys and routes are routed on. | none |
| models.catalog | List the routable models with their provider model, credential and source. | none |
| models.list | List the chassis model catalog — the valid targets for a route. | none |
| observability.metrics | Read daily request, token, cost and latency trends from Langfuse over a window (default: 14 days). | none |
| observability.traces.list | List Langfuse traces (names, timings, costs, tags), filtered by window, user, tag or name. Never a body. | none |
| packs.list | List the context packs: access policy, published and retired versions with their hashes, assignments and the settings. Never a version's text or a draft. | none |
| prices.list | List the rates the gateway prices each request with, per model, and where each rate came from. | none |
| providers.list | List the providers and models the chassis registry knows, for choosing a model to add. | none |
| reports.agent-metrics | Report the usage the agents reported themselves, per client, model, person, session, key or user. Never spend. | none |
| reports.developers | Report each developer's Claude Code usage from the client's own telemetry. Estimates, never spend. | none |
| reports.economics | Read cost economics: spend per client, model and tool, with the driver decomposition and the subscription lane. | none |
| reports.executive | Read the executive summary: spend, traffic, latency, top routes and models, and credit alerts. | none |
| reports.hooks | Report what the managed Claude Code hooks decided: denials, unscreened work and screening outages. | none |
| reports.list | List the reports the caller may open: built-in, their own, and shared with their role. | none |
| reports.providers | Report spend per provider credential, against its declared limit, with live rate-limit headroom. | none |
| reports.read | Read one report's definition: a built-in id, or a saved report's uuid the caller may see. The question, never the figures. | none |
| reports.run | Run one report (a built-in or saved report by id, or a draft spec) for the caller. | none |
| reports.security | Report security events: refused and failed requests, and the security-relevant audit entries. | none |
| reports.team | Report usage per person, rolled up through key ownership, with an unassigned bucket. | none |
| reports.tool-calls | Report tool calls through /mcp per tool, server, person or key: calls, outcomes, refusals, cost and latency. | none |
| retention.read | Read the retention policy (days kept per table; 0 = forever) and the last purge pass. | none |
| skills.list | List the company skills: who may select each, and every published or retired version's title, digest and declared requirements. Never a procedure's text, a draft or a review comment. | none |
| status.read | Read each catalog model's health: up, degraded, down or unknown, with recent traffic, errors and latency. | none |
| subscription-lane.read | Read whether a claude.ai subscription credential may cross the gateway, and what each mode does. | none |
| teams.list | List the teams a key may be put in, each with its tool grant and its member count. | none |
| tools.calls.list | List tool-call ledger rows, newest first: person, tool, decision, outcome, latency and cost, refused calls included. Never arguments or results. | none |
| tools.groups.list | List the tool groups: named bundles of servers, tools, resources and prompts that a grant may name. | none |
| tools.servers.inspect | Inspect one registered server from stored facts only: which grants reach it, its recent ledger activity and its name collisions. | none |
| tools.servers.list | List the registered MCP and REST servers with every tool, resource and prompt they declare. Never a credential. | none |
| tools.variables.read | Read a service-account key's server variables (a site, a workspace id), per server that declares them. Never a secret. | none |
| tools.virtual-servers.list | List the virtual servers: named toolsets an agent mounts as one endpoint. A mount only narrows a grant. | none |
| usage.report | Aggregate request usage grouped by one dimension over an optional date range. | none |
| usage.series | Read usage per hour or day: requests, tokens, cost, and latency and time-to-first-token percentiles. | none |
| usage.spend-split | Read interactive and batch spend for one window, side by side. Only the money is added up. | none |
| users.directory | List the accounts that can own a key: id, name and email only. Disabled accounts are left out. | none |
| users.list | List console users with their role, status, identity provider and last sign-in. | none |
| alerts.check | Evaluate every budget against the alert bands now, and deliver each new crossing once. Safe to repeat. | confirm |
| config.route.delete | Delete one route and its utterance examples. | confirm |
| config.route.upsert | Create or update one route (model, fallbacks, params, enabled). | confirm |
| config.settings.update | Update the default route and/or the classifier settings. | confirm |
| config.utterances.set | Replace one route's classifier examples. The gateway rebuilds its embedding index. 404 for an unknown route. | confirm |
| guardrails.test | Screen a sample against the rules for a route, key and tags, storing nothing. Budgeted per caller (429); each trial is audited without its text. | confirm |
| reports.save | Save a report for the caller: create one, or replace one they own. It stores the question, never the figures. | confirm |