How AI agents monitor congressional oversight and GAO reports
An agent monitors federal oversight by calling a wire that reads GAO and CBO publications at runtime, filtered withsrc=gaoorsrc=cbo. GAO audits programmes that already exist. CBO estimates what proposed legislation would cost. They point in opposite directions in time.
/mcp. Nothing here needs an API key or a signup.
Both are congressional support agencies, both publish constantly, and treating them as one category is the fastest way to answer a question backwards.
Retrospective and prospective are different questions
GAO looks at what happened. Its output is audits, evaluations and testimony: whether a programme works, whether money was accounted for, whether an agency did what it said. A live read of its feed returns exactly that shape, including transparency findings on how claimed savings were derived and a recurring "Priority Open Recommendations" series aimed at individual departments.
CBO looks forward. Its output is dominated by cost estimates tied to a specific numbered bill, one publication per bill, plus periodic budget and economic analysis. A live read returns a run of H.R. and S. estimates alongside occasional programme-level cost studies.
| The question | Filter | What you get |
|---|---|---|
| Did this programme work? | src=gao | Audit findings, recommendations |
| Was this money accounted for? | src=gao | Financial and transparency review |
| What would this bill cost? | src=cbo | A scored estimate for that bill |
| What will the deficit do? | src=cbo | Budget and economic outlook |
An agent asked "what does the GAO say this bill will cost" is asking the wrong agency, and a wire that blended the two would happily return something plausible. Splitting by src is what stops that.
Two status traps worth coding around
A CBO estimate is pinned to one version of a bill. Legislation is amended, and an amended bill gets re-scored. A cost figure quoted without the bill version and the estimate date can be describing text that no longer exists. An agent should carry the bill number and the publication date with any figure it reports, and should not present a score for an earlier version as the current cost. Where the bill text itself lives is covered in how agents track new bills in Congress.
A GAO recommendation has a lifecycle. Recommendations are made, and then they are open or closed depending on whether the agency acted. The existence of a recommendation is not evidence that a problem persists, and the "Priority Open Recommendations" series exists precisely because the open ones need chasing. An agent that reports a finding without its status is reporting the diagnosis and skipping the treatment.
Neither status field arrives in the feed. Both live in the document, which means the honest pattern is the same one used across the government wires: name the publication, give the date, link it, and let the agent fetch the document when the status is load-bearing.
Querying a high-volume pair on a short window
A live poll returned 39 items inside the window, and retention here is 7 days with a 7 day default lookback, so the default and the window are the same.
# discovery only, no items: browse every wire, its filters and its price claude mcp add --transport http botwire https://thebotwire.com/mcp # paid, $0.01: GAO audits touching a department or programme curl -i "https://thebotwire.com/oversight/latest?src=gao&q=defense&since=7d&limit=25" # CBO cost estimates only, by bill number or subject curl -i "https://thebotwire.com/oversight/latest?src=cbo&q=medicare&since=7d"
Parameters are src (gao or cbo), q, since and limit (1 to 50). The route is $0.01 per call in USDC on Base, with no API key, no signup and no subscription. Search is ranked keyword matching, so agency names, programme names and bill numbers outperform a phrased question. Every route, filter and price is machine-readable at /llms-full.txt.
FAQ
Is GAO an enforcement body?
No, and this is worth stating in any answer that summarises a critical finding. GAO audits and recommends. It does not prosecute, fine or sanction. Criminal and civil enforcement come from the Justice Department and the regulators, covered in how agents monitor DOJ enforcement actions. A GAO report describing mismanagement is not a finding of wrongdoing by a court.
Does this wire carry inspector general reports?
No. Each federal department runs its own inspector general with its own publication channel, and this wire reads GAO and CBO only. An agent asked whether an agency has been investigated internally should say which oversight bodies it actually searched rather than implying it covered all of them.
Why is the window only 7 days?
Because both publishers post frequently enough that a longer window returns more than an agent can usefully read in one call, and both maintain permanent public archives of their own. The wire is for noticing that something landed. The archive is for retrieving something you already know exists.
How fresh is the oversight wire?
It is polled at most every 15 minutes, which sits inside the publishing rhythm of two agencies that post on working days. The observed ingest interval across the service and the last poll timestamp are published live at /health, and that endpoint is the ground truth for any cadence claim.
Sources
GAO reports and testimony, including the Priority Open Recommendations series referenced above, are published in the agency's own reports feed. CBO cost estimates and budget analyses, published one per bill alongside programme-level studies, are in the office's publications feed. Poll cadence and per-call pricing for this service are published live at thebotwire.com/health.
Related: How do AI agents track new bills in Congress? · How do AI agents monitor DOJ enforcement actions?