Know when competitor prices and plan names change. Keep before-and-after evidence for your next pricing decision, sales conversation, or client report.
See the prices and plans we can detect on a public pricing page, then keep watching it with a free monitor.
Connect the remote MCP server and ask the agent to capture a free source baseline. The tool tells the agent what was observed, how often to check it, and when persistent monitoring is worth activating.
Free preflight checks work without an API key.
No checkout or API key is needed for the baseline.
The client should call this tool with this argument.
Response evidence varies by run.
Ground Truth sits between an agent’s discovery step and its decision to trust or act.
The agent finds an external URL, API endpoint, or claim.
Ground Truth fetches the resource and inspects observable response and header signals.
The agent receives PASS, WARN, or FAIL with structured evidence.
The agent continues, asks for caution, or stops before relying on the resource.
Start with the checks agents already use most: endpoint behavior, HTTP security configuration, and evidence for claims.
Check whether an unfamiliar resource responds, what status and content type it returns, whether it redirects, and whether auth or rate limits are blocking it.
check_endpointSee which common browser-facing headers are present or missing. The result is a configuration signal, not a vulnerability scan.
inspect_security_headersCheck supplied evidence URLs for the keywords you specify before an agent repeats a pricing, policy, support, or compliance claim.
verify_claimRun endpoint and header checks together and give the agent one decision-oriented result with both evidence sets.
preflight_endpointAn external resource can exist and still be the wrong thing to trust. Preflight makes the observable uncertainty explicit before the agent acts.
An endpoint can move, redirect, require credentials, rate-limit, or fail after an agent has discovered it.
Missing HSTS, CSP, or framing controls should be visible to the agent before it makes a browser-facing recommendation.
Keyword support on supplied public sources helps an agent show what it checked without pretending that silence proves a claim false.
PASS means continue, WARN means inspect the evidence, and FAIL means stop or ask for a different resource.
The agent should call the relevant tool as soon as it discovers an unfamiliar resource, not after it has already trusted the result.
Agent finds https://api.example.com/v1/users and calls preflight_endpoint before using it.
inspect_security_headers reports HTTPS and the headers it can observe, including missing recommendations.
verify_claim checks supplied policy or pricing pages for explicit supporting keywords.
The agent uses PASS, WARN, or FAIL to proceed, pause for review, or stop.
Create three persistent monitors before paying. Upgrade self-serve when your team needs hourly checks, more sources, or higher run volume.
Three real monitors with daily or weekly evidence checks.
check_endpointinspect_security_headerspreflight_endpointverify_claim (5 calls/month)For agents that need authoritative sources kept fresh automatically. Three tiers — 10, 40, or 150 active monitors.
preview_monitor first, then create a persistent monitor with a free watch key. Paid plans add hourly schedules, more sources, and larger quotas; x402 remains available for autonomous one-off research. Open the watch dashboard →
Use Ground Truth directly over HTTP if you want the free first check in a script, backend, or agent loop.
curl callInitialize the MCP session, then call the free endpoint check with no API key.
fetch exampleRun the same no-key endpoint check from code.
MCP stands for Model Context Protocol. Ground Truth is a preflight service for unfamiliar external resources. Use the direct server for free checks and Stripe API-key access; use the separate xpay deployment for pay-per-call clients.
Add Ground Truth to ~/Library/Application Support/Claude/claude_desktop_config.json. No key is needed for the free first call.
{
"mcpServers": {
"ground-truth": {
"url": "https://ground-truth-mcp.anishdasmail.workers.dev/mcp"
}
}
}
Add Ground Truth to .cursor/mcp.json in your project or ~/.cursor/mcp.json globally. Add X-API-Key only for team-plan paid tools.
{
"mcpServers": {
"ground-truth": {
"url": "https://ground-truth-mcp.anishdasmail.workers.dev/mcp"
}
}
}
PASS means the observed checks succeeded, WARN means the agent should inspect evidence or handle uncertainty, and FAIL means stop or choose another resource. None of these verdicts is a security guarantee.
Ground Truth fits anywhere an AI agent needs a final check before sharing an answer or taking action.
Verify pricing claims, confirm support entitlements, and check whether an API endpoint a customer asks about actually exists.
Test market assumptions, check whether a competitor exists, and compare live pricing pages before you lock in a direction.
Scan trust pages for SOC 2, ISO 27001, HIPAA, GDPR, DPA, SSO, and SCIM signals before repeating them to buyers or internal stakeholders.
Inspect public security headers, verify claims against public evidence, and add lightweight diligence before an agent recommends or approves a vendor.