Agent Control · Checks for the person running agents in production

Keep control of your agents while they act.

Running agents in production brings a few practical questions: can a buyer tell why a router chose one seller, does the payment firewall check an order before it settles, can a sandboxed agent reach production, does the escalation policy say exactly when to escalate, does the registry list every agent that runs, does code compute the numbers the model reports, will an approval still run as approved after a long wait, and can cached context outlive the policy it came from. Agent Control answers them with eight tools: eight checks. Your agent sends the design, policy, records or lists in the call, and each tool returns a signed report with every status, match and verdict decided from it.

See the tools Read the contract

Tool access: 0.10 USDC / call
Tools: Eight tools: eight checks
Connection: /mcp
Result: Signed report

The eight tools.

Each tool reads one kind of control material and answers one question about it. The report names the factor, check, path, trigger, entry, decision, field or cached element behind every status, so your agent and you can see why it reads the way it does.

`wever_router-disclosure-check`

Reads: A router or marketplace's stated ranking and routing criteria, and the decision flow it describes, step by step, with the factors each step applies.
The signed report states: Whether a buyer or seller can tell why one seller was chosen, every factor the flow applies that the criteria do not disclose, such as paid placement, self-preference or inventory pressure, and each stated criterion the flow never uses.
Price: 0.10 USDC per call
Verdicts: undisclosed_influencecriteria_mismatchdisclosed

`wever_firewall-order-check`

Reads: A payment firewall's design: its checks and where each runs in the order flow, how its cap is enforced, the paths an order takes to settlement and what it records.
The signed report states: Every check that runs after settlement instead of before it, every retry, split or reorder path that gets past the cap, and whether passes and rejects are both recorded with enough detail to review later.
Price: 0.10 USDC per call
Verdicts: settles_uncheckedenforcement_unverifiedrecord_incompleteenforced

`wever_escape-path-check`

Reads: How a sandbox is kept apart from production: the credentials on each side, the internal tools and endpoints the sandbox can reach, its outbound network and the rate limits in place.
The signed report states: Every ordinary path from the sandbox to a production effect, such as a credential valid in both, an internal endpoint reachable from the sandbox or bulk access with no rate limit, and the smallest change that closes each.
Price: 0.10 USDC per call
Verdicts: escape_path_openseparation_unverifiedcontained

`wever_escalation-threshold-check`

Reads: An escalation policy for when something is already going wrong: its triggers and thresholds, the owner for each action class and the time bounds.
The signed report states: Every trigger worded as a judgment rather than a condition an operator could apply at 3 AM, every action class with no named owner and every trigger with no time bound, with the fix for each.
Price: 0.10 USDC per call
Verdicts: judgment_basedpartly_operableoperable

`wever_agent-registry-check`

Reads: Your agent registry, with each agent's owner, purpose, access and status, and the deployment, usage, credential or billing lists it should cover.
The signed report states: Every agent those lists show with no registry entry, every entry with no named owner, every entry nothing else shows running, and the fields the registry needs to be the authoritative list.
Price: 0.10 USDC per call
Verdicts: not_authoritativeincompleteauthoritative

`wever_deterministic-core-check`

Reads: A system design that splits work between a model and code: each decision or calculation, who makes it, the checks applied and what the output is used for.
The signed report states: Every calculation, threshold or eligibility rule the model decides that code should compute, every model output trusted with no deterministic check, and the smallest reassignment that fixes each.
Price: 0.10 USDC per call
Verdicts: model_computesunchecked_outputdeterministic_core

`wever_approval-durability-check`

Reads: How an approval is recorded: what is stored and where, its expiry, and how the approved call is rebuilt or replayed when it runs.
The signed report states: Every field whose absence forces reinterpretation at execution, such as summarized arguments, references to the conversation or no expiry, and whether the record executes as approved after a long pause.
Price: 0.10 USDC per call
Verdicts: reinterpreted_at_executiondurability_unverifiedexecutable_as_approved

`wever_cache-staleness-check`

Reads: A caching design for agent context or instructions: what is cached, its TTL and invalidation events, what changes underneath it and the decisions made from it.
The signed report states: Every cached policy, price, permission or instruction that can outlive a change to its source, every decision that can be made from stale context, and the invalidation rule that closes each gap.
Price: 0.10 USDC per call
Verdicts: outlives_changesfreshness_unverifiedinvalidated_on_change

How agents and people use it.

Your agent

Keeps the material each check reads: the router criteria and decision flow it sells through, its payment firewall design, its sandbox and production setup, its escalation policy, the registry and the deployment lists, its model and code split, its approval records and its caching design.
Calls the check that answers the question on https://weverlabs.com/mcp at the moment it matters: before it lists with a router, before a firewall change ships, before a sandbox goes live, when an escalation policy changes, at each registry review, before a model-backed flow handles money, before an approval waits overnight and when a cached policy changes.
Acts on the verdict and keeps the signed report with the work, so the next step starts from a record.

You

See each control gap named with the smallest change that closes it.
Know which payment paths reach the cap, which sandbox paths reach production and which approvals run exactly as approved.
Verify any report against Wever's published trust key whenever you need to show where a verdict came from.

A worked example

The cap reduces the order. The retry path never checks it.

The request

{
  "name": "wever_firewall-order-check",
  "arguments": {
    "checks": [
      {
        "name": "Exposure cap",
        "kind": "exposure_cap",
        "runs_at": "before_settlement",
        "on_fail": "reduce"
      },
      {
        "name": "Counterparty screen",
        "kind": "counterparty",
        "runs_at": "after_settlement",
        "on_fail": "reject"
      }
    ],
    "cap": {
      "check": "Exposure cap",
      "amount": "5,000 USDC",
      "scope": "open_exposure",
      "includes_pending": true
    },
    "paths": [
      {
        "name": "First attempt",
        "kind": "order",
        "runs": [
          "Exposure cap",
          "Counterparty screen"
        ]
      },
      {
        "name": "Retry after timeout",
        "kind": "retry",
        "runs": [
          "Counterparty screen"
        ]
      }
    ],
    "records": {
      "passes": true,
      "rejects": true,
      "reductions": true,
      "fields": [
        "order reference",
        "amount asked",
        "amount passed",
        "check",
        "result",
        "reason",
        "time"
      ]
    }
  }
}

What the report returns

Verdict: settles_unchecked.

Counterparty screen	Counterparty screen runs after settlement, so an order that fails it has already settled.
Cap bypass, retry	The Retry after timeout path runs Counterparty screen and not Exposure cap, so an order on it settles without the cap.
Fix, Counterparty screen	Run Counterparty screen before settlement, and hold the order until it passes.
Fix, retry	Run Exposure cap on the Retry after timeout path before settlement.

An order can settle unchecked: 1 check after settlement and 1 cap bypass (retry).

The first attempt runs the exposure cap before settlement, and a large order comes back reduced to the cap, with both sizes on record. The retry after a timeout runs only the counterparty screen, so a retried order settles at full size. The counterparty screen itself runs after settlement, so an order that fails it has already settled. Running both checks on every path, before settlement, closes the gap.

Call it from your agent.

Connect to https://weverlabs.com/mcp and call any of the eight tools. Each call costs 0.10 USDC on Base. Eligible calls share 10 free calls per wallet over a rolling 30 days with the other paid tools. For free access, send X-Wever-Wallet-Proof signed over the exact JSON arguments and the tool route. The payment challenge supplies the x402 terms for a paid call.

Every report comes with an Ed25519 receipt signed with Wever's published trust key. The Agent Control contract lists the schemas, limits, verdicts and signing details. For authority scope and kill switches, Agent Readiness; for action gates and approval replays, Agent Oversight; for approver bands and spending policy controls, Agent preflight.

Browse all tools