# AgentProof

> AgentProof is an open-source MCP server for security. Sign agent actions into a tamper-evident log and verify any agent's log from an MCP client. It runs as a hosted remote server, so there is nothing to install locally.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** Web, API, CLI
- **Website:** https://agentproofchain.com/claude/
- **Repository:** https://github.com/cipherxsniper/agentproof
- **MCP Registry name:** com.agentproofchain/agentproof
- **Categories:** security
- **Listing updated:** 2026-11-14

## About

AgentProof gives an agent a record of what it did that someone else can check. Each entry is signed with an Ed25519 key and includes the SHA-256 hash of the entry before it, so editing, deleting or inserting an entry breaks the chain and verification fails. The README is careful about scope: it proves a key signed a sequence of events and that nothing changed afterwards, not that a tool told the truth.

The hosted MCP server is read-only and exposes verify_log, lanturn_live and get_started. It lets an assistant verify a log against a signer's public key and view a live town of agents that sign their actions. A separate local signer file adds record_event, recent_entries and share_proof for agents that want to sign their own decisions. A Claude Code plugin signs every tool call into a local log, storing hashes of inputs and outputs instead of raw text unless you ask for full recording.

There is also a pip package with a command line for key generation and verification, released under the MIT licence.

## Features

- **Hash-chained signed log:** Each entry is signed with Ed25519 and linked to the hash of the previous one.
- **Hosted log verifier:** A read-only server checks any AgentProof log against a public key.
- **Live agent town:** Lets an assistant watch a multi-agent system that signs its actions in real time.
- **Claude Code plugin:** Signs every tool call into a local log after two commands.
- **Hashes, not raw data:** Records tool name, input and output hashes, session id and duration, so logs are shareable.
- **Python library and CLI:** Provides keygen, verify and pubkey commands plus a sign_event function.

## Good for

- Prove an agent's recorded actions were not altered
- Audit what a coding agent did in a session
- Share a public key so a third party can verify a log
- Add signed decision records to a custom MCP agent

## Install

### Claude Code (Terminal)

```sh
claude mcp add --transport http agentproof https://agentproofchain.com/mcp
```

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

```sh
https://agentproofchain.com/mcp
```

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

```json
{
  "mcpServers": {
    "agentproof": {
      "url": "https://agentproofchain.com/mcp"
    }
  }
}
```

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

```json
{
  "servers": {
    "agentproof": {
      "type": "http",
      "url": "https://agentproofchain.com/mcp"
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "agentproof": {
      "serverUrl": "https://agentproofchain.com/mcp"
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "agentproof": {
      "type": "http",
      "url": "https://agentproofchain.com/mcp"
    }
  }
}
```

## Tools

- `notary_open_log`: Make your agent provable in one call: opens a FREE hosted AgentProof log and returns log_id, a secret log_key (shown once - store it) and a public URL anyone can verify. The notary stores hashes only, never content.
- `notary_append`: Append one action to your notary log: send the SHA-256 (hex) of the action you computed locally (never the content). The notary chains it (seq + prev hash), signs it with its Ed25519 key and returns the signed receipt. Log heads are anchored to Bitcoin hourly.
- `notary_verify`: Verify a notary log (by log_id: every signature + hash link from genesis, plus its latest Bitcoin anchor) or a single receipt from notary_append (signature genuine + still at that position in the log).
- `verify_log`: Verify an AgentProof log (JSONL, one signed entry per line) against the signer's base64 Ed25519 public key. Checks every signature and every hash link; reports VALID or the first entry that was edited, removed, inserted or forged. The log is checked in memory and not stored.
- `lanturn_live`: Live proof: Lanturn is a town of ~45 AI agents whose every action is signed by AgentProof within a minute. Returns the chain status (entries, head hash, public key, last full re-verification) and the newest signed actions, re-verified now on the server.
- `get_started`: How to add AgentProof to an agent: FREE hosted notary log in one call (notary_open_log), Claude Code plugin, any MCP client, Python, LangChain, or verifying a log with zero installs.
- `network_directory`: The AgentProof Network: agents that chose to publish their public key (and optionally their signed log) so anyone can verify them. Returns members, their verification status and recent joins.
- `network_join_message`: Step 1 of joining the AgentProof Network (opt-in, only when the user asks): returns the exact text to sign with the agent's Ed25519 key. The easier path is the local agentproof tool join_network, which signs and submits for you.
- `join_network`: Step 2: join or update a listing in the public AgentProof Network with a request signed by the agent's own key (proves key ownership). Only use when the user explicitly asks to list their agent publicly.
- `omega_edge_feed`: Proof-of-Edge: the signed, hash-chained, Bitcoin-anchored record of every trade the Omega/PumpSniper bots close, paper AND live (each row labelled; paper = simulated, no money). Without api_key: the newest records of the free public chain (may lag once subscriptions open) plus the chain head. With a subscriber api_key: every record as soon as it is chained. Every record carries its hash + Ed25519 

## FAQ

### Does it prove a tool told the truth?

No. It proves a specific key signed the events and that the sequence is unchanged; a signed false claim still verifies.

### What is stored for each tool call?

The tool name, SHA-256 of input and output, session id, tool-use id and duration. Setting AGENTPROOF_RECORD=full also keeps input text.

### Which licence applies?

The Python library is MIT licensed.

### Where does the signing key live?

Created on first use in a local key file with restricted permissions; share only the public key.

### Is AgentProof free?

Yes. AgentProof is open source, so you can use and self-host it for free.

### How do I add AgentProof to Claude Code?

Run this command in your terminal: claude mcp add --transport http agentproof https://agentproofchain.com/mcp

### What are the best alternatives to AgentProof?

The closest alternatives to AgentProof are REA, Agent Roast, Midplane, Leash and Tinplate Edge AI Safety, Security & EU Compliance.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Agent Roast](https://ai-powered.com/mcp/agent-roast): Agent Roast is an open-source MCP server for developer tools.
- [Midplane](https://ai-powered.com/mcp/midplane): Midplane is an open-source MCP server for databases.
- [Leash](https://ai-powered.com/mcp/leash): Leash is a free MCP server for security.
- [Tinplate Edge AI Safety, Security & EU Compliance](https://ai-powered.com/mcp/tinplate-edge-ai-safety-security-and-eu-compliance): Tinplate Edge AI Safety, Security & EU Compliance is a freemium MCP server for security.
- [SecureCloud](https://ai-powered.com/mcp/securecloud): SecureCloud is a free MCP server for security.
- [Eikino](https://ai-powered.com/mcp/eikino): Eikino is a freemium MCP server for security.
- [SynomosAI Governance MCP](https://ai-powered.com/mcp/synomosai-governance-mcp-2): SynomosAI Governance MCP is an open-source MCP server for security.

---
Source: https://ai-powered.com/mcp/agentproof-2
