AgentLog-MCP
An open-source project that achieves tamper-proof verification of operational logs by adding cryptographic signatures to each execution step of an AI agent. It supports integration with any compatible client for auditing purposes via the MCP protocol.
Timeline (last 90 days)
AgentLog-MCP is an open-source project that signs every step executed by an AI agent and allows anyone to verify whether the log records have been tampered with by asking questions in natural language.
The underlying component of AgentLog-MCP is an open-source library called agent-log, released under the MIT license and accompanied by 56 test cases.
The verification conclusion of AgentLog-MCP is not derived by reading a status flag bit, but is recalculated in real time from hash values and signatures, so tampering with a flag bit cannot pass the check.
In adversarial tests, regardless of whether an attacker modifies log content, re-signs with an intruder's key, or deletes a step, AgentLog-MCP's cryptographic verification is able to detect the anomaly.
In dispute scenarios between clients and operators, AgentLog-MCP arbitrates based on the agent-signed version, and the copy whose signature matches is determined to be authentic.
AgentLog-MCP adopts an on-chain anchoring and off-chain storage architecture: raw logs are stored in a traditional database, hash digests are written to a public blockchain testnet, and anyone can read and verify them on-site via RPC without needing a key.
The wire panel in the demo displays every MCP protocol message, including the initialize handshake (protocol version November 25, 2025), MCP apps extensions, tools list, and tools call.
The AgentLog-MCP project itself was built with the assistance of AI coding agents such as Claude Code and has undergone code review.
The problem with traditional logging is that the recorder and the auditee are often the same party, lacking independence, whereas AgentLog-MCP decouples the authenticity of records from their storage location through cryptographic signatures.
AgentLog-MCP is currently still a demo and an early-stage open-source library, and there is still a verification process before it reaches production-grade reliability.
All Facts (10)
AgentLog-MCP is an open-source project that signs every step executed by an AI agent and allows anyone to verify whether the log records have been tampered with by asking questions in natural language.
50%UnverifiedThe underlying component of AgentLog-MCP is an open-source library called agent-log, released under the MIT license and accompanied by 56 test cases.
50%UnverifiedThe verification conclusion of AgentLog-MCP is not derived by reading a status flag bit, but is recalculated in real time from hash values and signatures, so tampering with a flag bit cannot pass the check.
50%UnverifiedIn adversarial tests, regardless of whether an attacker modifies log content, re-signs with an intruder's key, or deletes a step, AgentLog-MCP's cryptographic verification is able to detect the anomaly.
50%UnverifiedIn dispute scenarios between clients and operators, AgentLog-MCP arbitrates based on the agent-signed version, and the copy whose signature matches is determined to be authentic.
50%UnverifiedAgentLog-MCP adopts an on-chain anchoring and off-chain storage architecture: raw logs are stored in a traditional database, hash digests are written to a public blockchain testnet, and anyone can read and verify them on-site via RPC without needing a key.
50%UnverifiedThe wire panel in the demo displays every MCP protocol message, including the initialize handshake (protocol version November 25, 2025), MCP apps extensions, tools list, and tools call.
50%UnverifiedThe AgentLog-MCP project itself was built with the assistance of AI coding agents such as Claude Code and has undergone code review.
50%UnverifiedThe problem with traditional logging is that the recorder and the auditee are often the same party, lacking independence, whereas AgentLog-MCP decouples the authenticity of records from their storage location through cryptographic signatures.
50%UnverifiedAgentLog-MCP is currently still a demo and an early-stage open-source library, and there is still a verification process before it reaches production-grade reliability.
50%