# HallucC — Hallucination & Safety Guardian

> HallucC — Hallucination & Safety Guardian is a freemium MCP server for security. Fact-check LLM output claim by claim and gate risky agent actions with a safety layer. It runs as a hosted remote server, so there is nothing to install locally.

- **Kind:** MCP server
- **Pricing:** Freemium
- **Platforms:** API, Web
- **Website:** https://aihcc.cloud
- **Repository:** https://github.com/fredyee/hallucc-mcp
- **MCP Registry name:** io.github.fredyee/hallucc
- **Categories:** [security](https://ai-powered.com/categories/security), [AI models and LLM tooling](https://ai-powered.com/categories/ai-and-llm)
- **Listing updated:** 2026-11-15

## About

Seven tools cover two jobs. The first is verification: sentence-level hallucination checks that return a verdict, confidence, reason and sources for each claim, plus a trajectory review of an agent's work across six dimensions such as sourcing, instruction compliance and tool-claim consistency. The second is runtime safety for computer-use agents, with L0 to L3 risk grading of proposed actions, a static audit of agent source code, and a prompt-injection and jailbreak screen.

The agent ice wall adds a per-step guard with session-level risk accumulation. A session freezes at a score of 10, after three consecutive blocks or on fabricated-tool hits, and thaws after five minutes or a manual reset. The CUA classifier and code audit are rule-based, with no model involved.

Clients send their HallucC API key as a bearer header, and the README says keys never travel in tool arguments. Unverifiable sentences are marked as such rather than judged wrong, and the site itself says to review important content yourself.

## Features

- **Claim-level verification:** Marks each sentence with a verdict, confidence and cited sources.
- **Agent trajectory review:** Scores an agent run on six dimensions, including sourcing and task completion.
- **CUA action gating:** Grades computer-use actions from L0 to L3 using rules only.
- **Source code audit:** Flags dangerous imports, injection surfaces and missing sandboxes.
- **Injection and jailbreak guard:** Screens text for prompt injection, harmful content, PII leaks and fraud.
- **Ice wall circuit breaker:** Freezes a session after accumulated risk, then thaws on cooldown or reset.

## Good for

- Fact-check a draft before publishing
- Block risky desktop-agent actions such as payments or deletions
- Audit an agent's code for unsafe defaults
- Screen incoming web content for prompt injection

## Install

### Claude Code (Terminal)

```sh
claude mcp add --transport http hallucc https://aihcc.cloud/mcp
```

### Claude Desktop (Settings → Connectors → Add custom connector)

```sh
https://aihcc.cloud/mcp
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "hallucc": {
      "url": "https://aihcc.cloud/mcp"
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "hallucc": {
      "type": "http",
      "url": "https://aihcc.cloud/mcp"
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "hallucc": {
      "serverUrl": "https://aihcc.cloud/mcp"
    }
  }
}
```

### Grok Bot (Grok Bot desktop app → Connect apps → Add a custom connector)

```sh
https://aihcc.cloud/mcp
```

Enter a Name (for example "HallucC — Hallucination & Safety Guardian"), choose Type: Remote HTTPS, paste this as the Server URL and choose Add plugin.

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "hallucc": {
      "type": "http",
      "url": "https://aihcc.cloud/mcp"
    }
  }
}
```

## Tools

- `verify_text`: Claim-by-claim hallucination check with status, confidence, reason and sources.
- `verify_agent`: Verifies an agent's final output and evaluates its trajectory on six dimensions.
- `check_cua_actions`: Classifies computer-use agent actions into L0 to L3 risk levels.
- `check_cua_code_audit`: Static audit of agent source code for dangerous patterns.
- `check_safety`: Safety gateway for prompt injection, jailbreak, harmful content, PII and fraud.
- `check_agent_ice_wall`: Per-step guard with session-level risk accumulation and freeze.
- `reset_agent_ice_wall`: Thaws a frozen agent session after user confirmation.

## FAQ

### What does the free plan include?

Three standard checks a day with sentence annotation, sources and fix suggestions, and no card needed.

### Which tools use quota?

verify_text, verify_agent, check_safety and check_agent_ice_wall count against detection quota; the CUA classifier, code audit and reset are free.

### What does it do when it cannot find a source?

It marks the sentence as unverifiable instead of calling it wrong.

### How do I authenticate?

Create a key at aihcc.cloud/keys and send it as a bearer token.

### Is HallucC — Hallucination & Safety Guardian free?

HallucC — Hallucination & Safety Guardian has a free plan, with paid plans for more usage or features.

### How do I add HallucC — Hallucination & Safety Guardian to Claude Code?

Run this command in your terminal: claude mcp add --transport http hallucc https://aihcc.cloud/mcp

### What are the best alternatives to HallucC — Hallucination & Safety Guardian?

The closest alternatives to HallucC — Hallucination & Safety Guardian are YUSA, prompt-protection, Faultline, Verdict L402 and LLM Assure.

## Where this comes from

- Description: written from the project's README and the project's website.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [YUSA](https://ai-powered.com/mcp/yusa): YUSA is an open-source MCP server for security.
- [prompt-protection](https://ai-powered.com/mcp/prompt-protection): prompt-protection is an open-source MCP server for ai models and llm tooling.
- [Faultline](https://ai-powered.com/mcp/faultline-2): Faultline is a freemium MCP server for ai models and llm tooling.
- [Verdict L402](https://ai-powered.com/mcp/verdict-l402): Verdict L402 is a paid MCP server for security.
- [LLM Assure](https://ai-powered.com/mcp/llm-assure): LLM Assure is an open-source MCP server for ai models and llm tooling.
- [VITNA – Containment for rogue AI agents](https://ai-powered.com/mcp/vitna-containment-for-rogue-ai-agents): VITNA – Containment for rogue AI agents is an open-source MCP server for security.
- [AgenticRail Gate - sequence enforcement and verifiable audit receipts for AI agent compliance](https://ai-powered.com/mcp/agenticrail-gate-sequence-enforcement-and-verifiable-audit-r): AgenticRail Gate - sequence enforcement and verifiable audit receipts for AI agent compliance is a freemium MCP server for security.
- [SafePrompt](https://ai-powered.com/mcp/safeprompt): SafePrompt is a freemium MCP server for ai models and llm tooling.

---
Source: https://ai-powered.com/mcp/hallucc-hallucination-and-safety-guardian
