pic-standard

Overview

The pic-standard MCP server implements a safety protocol that constrains what AI agents are allowed to do based on verifiable provenance and evidence. It prevents prompt injections and blind tool calls by requiring agents to prove their intent, impact, provenance, and evidence before high-impact actions execute. The server works 100% locally with zero cloud dependencies and integrates with MCP to provide fail-closed tool guarding.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• verify

  Verify an action proposal against the PIC contract

  Try:Verify this action proposal for financial transferTry:Check if this tool call is allowed by PIC standardsTry:Verify the evidence for this high-impact action
• keys

  Manage the keyring of trusted signers

  Try:Show me the current trusted keysTry:Generate a new keyring exampleTry:Add a new trusted signer

Note: Inferred from CLI commands in the README. While the MCP integration is mentioned, specific MCP tool names aren't clearly documented in the provided excerpt.

Comparable tools

Installation

pip install pic-standard[mcp]

{
  "mcpServers": {
    "pic": {
      "command": "python",
      "args": ["-m", "pic_standard.mcp"]
    }
  }
}

FAQ

What is PIC?

PIC (Provenance & Intent Contracts) is a lightweight protocol that forces AI agents to prove every important action before it happens, requiring agents to declare intent, impact, provenance, and evidence.

How does PIC integrate with MCP?

PIC provides an MCP integration that offers enterprise tool guarding with fail-closed defaults, request correlation, DoS limits, and evidence sandboxing for MCP tools.

On Hacker News

Recent discussion from the developer community.

• Fail-closed evidence for LLM tool calls (SHA-256 and MCP)▲ 1
  Story by fabsalvadori · 2026-01-23
• Show HN: Pic-Standard – Open Protocol for Agentic AI Safety▲ 1
  Story by fabsalvadori · 2026-01-13

Compare pic-standard with