DCL Trust Oracle — Claude Code MCP server for output audits
Checks agent responses against policies and records verifiable decision metadata. Use it to screen outputs for injection attempts, unsafe text, or exposed secrets.
claude mcp add --transport http dcl-trust-oracle https://mcp.fronesislabs.com/mcpWhat DCL Trust Oracle — AI/LLM Output Audit (x402 MCP) does
DCL Trust Oracle evaluates agent output and adds hashes and decision metadata to a tamper-evident chain; it does not store raw message content. An audit lookup by transaction hash can return the verdict, confidence, reason, and agent details. The chain also makes changes to past records detectable.
Its MCP tools cover prompt injection, harmful text, secret and PII exposure, plus format adherence and contextual drift. You can submit a single result or a batch of up to 200 items per MCP call. After an action, request a basic or deep audit; a session tool supports ongoing drift tracking. Calls are paid, with prices starting at $0.01 per call.
Claude Code connects over HTTP to the hosted MCP endpoint, so local server startup is unnecessary. For local development, the project requires Python 3.11+, requirements.txt, and mcp_server.py. The deep forensic audit costs $0.50 per call, so choose it when its extra context is needed.
What it helps with
- screen an agent response before it reaches a user
- detect prompt injection and role hijacking attempts
- scan output for exposed credentials or PII
- inspect an earlier decision using its transaction hash
- track drift across a long-running audit session
Example prompts for the agent
- Check this response for prompt injection before it reaches the user.
- Scan this output for exposed credentials and personal information.
- Use the transaction hash to retrieve the audit and explain its verdict.
How to install DCL Trust Oracle — AI/LLM Output Audit (x402 MCP)
Platform CLI: remote server
claude mcp add --transport http dcl-trust-oracle https://mcp.fronesislabs.com/mcpPlatform CLI: remote server
codex mcp add dcl-trust-oracle-aillm-output-audit-x402-mcp --url https://mcp.fronesislabs.com/mcpConfig file
{
"mcpServers": {
"dcl-trust-oracle-aillm-output-audit-x402-mcp": {
"url": "https://mcp.fronesislabs.com/mcp"
}
}
}Config file
{
"servers": {
"dcl-trust-oracle-aillm-output-audit-x402-mcp": {
"type": "http",
"url": "https://mcp.fronesislabs.com/mcp"
}
}
}Platform CLI: remote server
gemini mcp add -t http dcl-trust-oracle-aillm-output-audit-x402-mcp https://mcp.fronesislabs.com/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about DCL Trust Oracle — AI/LLM Output Audit (x402 MCP)
How do I install DCL Trust Oracle in Codex?
Run codex mcp add dcl-trust-oracle-aillm-output-audit-x402-mcp --url https://mcp.fronesislabs.com/mcp. Codex connects to the hosted MCP endpoint over HTTP.
What does DCL Trust Oracle do?
It evaluates agent output against policies and records hashes plus decision metadata in a verifiable chain. You can retrieve an audit by transaction hash.
Can it check multiple outputs in one call?
Yes. The dcl_evaluate_batch tool accepts up to 200 items per MCP call.
