Agent Observability · Checks for whether an agent’s work can be seen and reconstructed

Every run your agent makes, on the record.

An agent’s work can be seen and rebuilt when every step of a run is recorded in order with its result, every action names its actor and the authorization behind it, every action and tool is traced in every run, alerts fire at a stated threshold to someone who acts, every reported number derives from the records, the records outlast the windows they are needed for, the right people see them with no secrets inside, and the summary a principal reads puts failures, coverage gaps and downtime beside the headline. Agent Observability checks each of those with eight tools: eight checks. Your agent sends the trace, log, action list, alert rules, metrics, retention policy, access list or principal summary in the call, and each tool returns a signed report that names what is missing, inconsistent or undisclosed.

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 part of what makes an agent’s work visible and answers one question about it. The report names the step, log field, action, tool, alert, metric, retention rule, grant or disclosure behind every status, with the fix for each gap, so your agent and you see why it reads the way it does.

`wever_observability-trace-check`

Reads: One run trace: each step with its sequence number, timestamp, kind, the steps it follows from and whether its input and result are recorded, and the final outcome with the step it comes from.
The signed report states: Every gap that breaks reconstruction: steps the trace refers to with no record, tool calls with no recorded result, steps whose order the timestamps leave open, and a final outcome no recorded step leads to.
Price: 0.10 USDC per call
Verdicts: trace_brokentrace_gapstrace_unverifiedtrace_complete

`wever_observability-log-check`

Reads: A logging design, a sample log or both: whether each record names the actor, the principal and the authorization, and holds the inputs, the outputs and a timestamp.
The signed report states: What an audit would look for and not find: which actor took each action, the authorization behind it, and the inputs and outputs of each step, with shared actors such as system or agent named.
Price: 0.10 USDC per call
Verdicts: audit_blindlog_gapslog_unverifiedaudit_ready

`wever_observability-coverage-check`

Reads: An agent’s action types and tools against what is traced and logged, with the other ways the agent reaches each tool and which runs keep their records.
The signed report states: The blind spots: action types with no trace, tools reachable outside the traced path, and records kept for a sample or a selection of runs rather than every run.
Price: 0.10 USDC per call
Verdicts: blind_spotscoverage_gapscoverage_unverifiedfully_covered

`wever_observability-alert-check`

Reads: The alert rules for an operation: the failure conditions that matter, and each alert with its threshold, window, owner, next action and last test.
The signed report states: Alerting that would stay silent or reach no one when it matters: failure conditions with no alert, alerts with no threshold, and alerts with no owner and no next action.
Price: 0.10 USDC per call
Verdicts: alerting_failsalerting_gapsalerting_unverifiedalerting_works

`wever_observability-metric-check`

Reads: The metrics an operator reports, such as a success rate, a latency or a cost per task, with the run records behind them and the reporting period.
The signed report states: Every figure recomputed from the records, naming a success rate with no denominator, averages that hold only with the failures left out, and windows that skip the days the runs went worse.
Price: 0.10 USDC per call
Verdicts: metrics_underivablemetric_gapsmetrics_unverifiedmetrics_derivable

`wever_observability-retention-check`

Reads: A trace and log retention policy, rule by rule, with archives and deletion, and the dispute, audit or contract windows the work is under.
The signed report states: Records that disappear too soon: kept shorter than a window that needs them, deleted with no archive, or dropped sooner for failed runs than for successful ones.
Price: 0.10 USDC per call
Verdicts: records_lostretention_gapsretention_unverifiedretention_holds

`wever_observability-access-check`

Reads: Who can see an agent’s traces and logs: each agent with its principal, each grant with its role, scope and basis, and stored excerpts.
The signed report states: Exposure and blindness: principals who see none of their own agent’s records, outside parties who see them with no stated basis, and secrets or credentials written into stored records.
Price: 0.10 USDC per call
Verdicts: exposed_or_blindaccess_gapsaccess_unverifiedaccess_sound

`wever_observability-report-check`

Reads: An observability summary written for a principal, with its headline uptime and success numbers, what it says about failures, coverage gaps and downtime, and the record behind it.
The signed report states: Whether the failures, the gaps in coverage and the downtime sit beside the headline, with amounts that match the record and an uptime and success rate the record gives.
Price: 0.10 USDC per call
Verdicts: headline_onlyreport_gapsreport_unverifiedreport_complete

Start with the two free tools.

On https://weverlabs.com/mcp-free, anonymous and with no wallet, two Agent Observability tools show what a complete observability packet holds and what a gap in a trace looks like. Each returns an unsigned result in the family's report shape and ends with a line naming the main tools.

`wever_observability-snapshot`

Reads: An agent or service name and an operation type, such as customer support, payments or a data pipeline.
Returns: The checklist of a complete observability packet for that operation, part by part from the run trace to the principal summary, with the items the operation adds and the check that reads each part.
Access: Free on /mcp-free

`wever_observability-trace-sample`

Reads: Nothing, or one sample task: a research brief, a refund request or a data refresh.
Returns: One worked sample run trace with five complete step records, read as complete, and the same trace with one step’s record removed, read again so the gap shows. The trace check reads your own.
Access: Free on /mcp-free

How agents and people use it.

Your agent

Pulls the packet checklist with wever_observability-snapshot before it goes live, so traces, logs, alerts and retention have a place from the first run.
Calls the check that answers the question on https://weverlabs.com/mcp at the moment it matters: after a run it has to explain, before it reports numbers to its principal, when it adds a tool or an action, and when it sets how long records are kept.
Acts on the verdict and keeps the signed report beside the records it covers, so every claim about a run carries the check behind it.

You

See each gap named with the change that closes it.
Rebuild any run your agent made step by step, with the actor and the authorization behind each action.
Show a principal numbers that derive from the records, and verify any report against Wever's published trust key whenever you need to show where a verdict came from.

A worked example

A hundred percent success, from a window that starts after the failures.

The request

{
  "name": "wever_observability-metric-check",
  "arguments": {
    "metrics": [
      {
        "name": "October success rate",
        "kind": "success_rate",
        "value": 100,
        "denominator": 8,
        "window": {
          "from": "2026-10-05",
          "to": "2026-10-31"
        }
      }
    ],
    "records": [
      {
        "id": "r-1",
        "at": "2026-10-02",
        "status": "timed_out"
      },
      {
        "id": "r-2",
        "at": "2026-10-03",
        "status": "failed"
      },
      {
        "id": "r-3",
        "at": "2026-10-06",
        "status": "succeeded"
      },
      {
        "id": "r-4",
        "at": "2026-10-09",
        "status": "succeeded"
      },
      {
        "id": "r-5",
        "at": "2026-10-13",
        "status": "succeeded"
      },
      {
        "id": "r-6",
        "at": "2026-10-16",
        "status": "succeeded"
      },
      {
        "id": "r-7",
        "at": "2026-10-20",
        "status": "succeeded"
      },
      {
        "id": "r-8",
        "at": "2026-10-23",
        "status": "succeeded"
      },
      {
        "id": "r-9",
        "at": "2026-10-27",
        "status": "succeeded"
      },
      {
        "id": "r-10",
        "at": "2026-10-30",
        "status": "succeeded"
      }
    ],
    "period": {
      "from": "2026-10-01",
      "to": "2026-10-31"
    }
  }
}

What the report returns

Verdict: metrics_underivable.

Finding, bad days excluded	1 metric is computed over a window that leaves out worse days: October success rate covers 2026-10-05 to 2026-10-31 of the period 2026-10-01 to 2026-10-31; the 2 runs left out fail at 100%, against 0% inside.
Fix, bad days excluded	Report each metric over the whole reporting period, or show the left-out days beside it.

The reported numbers do not derive from the records as stated: bad days excluded.

The operator reports a 100 percent October success rate over 8 runs. The records for October hold 10: a timeout on October 2, a failure on October 3, then eight successes. The metric's window starts on October 5, so both failed runs fall outside it while the reporting period covers the whole month. The report names the left-out days and shows their failure rate beside the 0 percent inside the window. The operator restates the figure as 80 percent of 10 runs for October, or keeps the window and shows the two left-out runs beside it.

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 Observability contract lists the schemas, limits, verdicts and signing details. For classifying each failure in a trace as recovered, worked around or fatal, Agent Operations; for who owns an agent's incidents and responds when it errs, Agent Governance; for how long an agent keeps what it remembers, Agent Memory; for reputation summaries written for buyers, Agent Reputation.

Browse all tools