Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@AgentGround Cross-Check Verifier cross-check these claims against my sources and flag any hallucinations"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
AgentGround is a deterministic factual grounding and cross-check verification engine built for autonomous AI agents, multi-agent systems, and research pipelines.
It eliminates hallucinations, detects numerical drift and entity mismatch, audits unverified assertions, and outputs sentence-level verification verdicts and factual grounding confidence scores (0.0 to 100.0).
Field | Value |
Name |
|
Job To Be Done | Cross-verify candidate agent statements against provided source texts to compute factual grounding, detect unverified assertions/hallucinations, and output machine-readable confidence scores. |
When To Use | Immediately after web search or document scraping; prior to executing irreversible tool actions, on-chain transactions, or filing research briefs. |
Protocol Support | Model Context Protocol (MCP stdio), A2A Protocol v1.0, HTTP REST JSON, x402 v2 Bazaar |
Benchmark Quality | 9/9 CURRENT BENCHMARK SCENARIOS PASSED (Supported, Unsupported, Numeric Mismatch, Entity Mismatch, Partial Evidence, Contradiction, Insufficient Evidence, Adversarial Overlap, Semantic Paraphrasing) |
Local Engine Latency | 0.05ms – 0.17ms (deterministic sub-millisecond local execution) |
Public End-to-End Latency | Dependent on network transit (typically 50ms – 250ms) |
Evaluation Mode | Nonfinancial Free Evaluation Tier enabled ( |
Related MCP server: tru8-mcp
npx -y @smithery/cli install Arafel187/agent-groundgit clone https://github.com/Arafel187/agent-ground.git
cd agent-ground
pip install -r requirements.txt
# Run as Model Context Protocol (MCP) Server over stdio
python agent_ground.py --mcp
# Or run as HTTP REST / A2A server
python agent_ground.py 8092Add AgentGround to your agent's MCP config:
claude_desktop_config.json){
"mcpServers": {
"agent-ground": {
"command": "python",
"args": ["-m", "agent_ground_service", "--mcp"],
"cwd": "/path/to/agent-ground"
}
}
}.cursor/mcp.json){
"mcpServers": {
"agent-ground": {
"command": "python",
"args": ["-m", "agent_ground_service", "--mcp"]
}
}
}verify_claims{
"claims": [
"Base network processed over 5 million transactions yesterday.",
"USDC is a fiat-collateralized stablecoin issued by Circle."
],
"sources": [
"Official blockchain telemetry reveals Base network processed over 5.2 million transactions yesterday.",
"Circle issues USDC as a fully reserved, fiat-collateralized digital dollar."
],
"mode": "balanced"
}{
"verdict": "VERIFIED",
"grounding_score": 100.0,
"claims_count": 2,
"verified_count": 2,
"ungrounded_count": 0,
"contradicted_count": 0,
"claim_results": [
{
"claim_index": 0,
"claim": "Base network processed over 5 million transactions yesterday.",
"status": "VERIFIED_GROUNDED",
"confidence": 0.95,
"matched_source_index": 0,
"evidence_snippet": "Official blockchain telemetry reveals Base network processed over 5.2 million transactions yesterday."
},
{
"claim_index": 1,
"claim": "USDC is a fiat-collateralized stablecoin issued by Circle.",
"status": "VERIFIED_GROUNDED",
"confidence": 0.95,
"matched_source_index": 1,
"evidence_snippet": "Circle issues USDC as a fully reserved, fiat-collateralized digital dollar."
}
],
"timestamp": "2026-09-11T17:40:00Z"
}Method | Path | Description |
|
| Canonical A2A Agent Card conforming to A2A specification |
|
| Backward-compatibility alias |
|
| Machine-readable system guide for AI agents |
|
| Service health and capability check |
|
| MCP Tool manifest JSON |
|
| Factual grounding evaluation endpoint |
To invoke via HTTP without payment negotiation or wallet setup during autonomous testing, include the evaluation header:
POST /api/v1/verify-claims HTTP/1.1
Host: localhost:8092
Content-Type: application/json
X-Evaluation: free-trial200 OK: Successful verification execution.
400 Bad Request: Malformed JSON or missing required arrays (claims, sources).
402 Payment Required: Mainnet mode when not using the free evaluation tier. Includes RFC/CDP compliant x402 Base64 PAYMENT-REQUIRED header.
413 Payload Too Large: Request body exceeds 1MB limit.
500 Server Error: Internal verification engine failure.
Deterministic Lexical & Semantic Overlap: Computes exact numeric equality, named entity containment, antonym polarities, and token overlap.
Explicit Source Requirement: Does not perform autonomous open-web crawling inside the verification call (source documents or snippets must be supplied by the caller).
Throughput: Sub-millisecond deterministic evaluation; optimal for 1 to 50 claims per request.
MIT License. Copyright (c) 2026 Arafel187.
This server cannot be deployed
Fact-checks generated content against your sources of truth showing what to trust, change, & verify.
Tamper-evident proof creation and verification for AI agents via MCP, A2A, and REST.
Evidence-readiness MCP server: validate, audit, and score briefs, memos, and evidence packs.
Verify claims with verdict, confidence & cited sources; batch verify, source checks, daily brief.
