# agentkitai/agentlens [Health: Active]

**Category:** 📊 Monitoring  
**Repository:** https://github.com/agentkitai/agentlens  
**GitHub Stars:** 23  
**npm Downloads (last month):** 349  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/agentkitai-agentlens

## Description
Tamper-evident observability for AI agents: a SHA-256 hash-chained audit log with chain verification and signed export (EU AI Act Art. 12). Instrument any agent with zero code via npx -y @agentlensai/mcp; also ingests OpenTelemetry GenAI traces.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "agentlens": {
    "command": "npx",
    "args": ["-y","@agentkitai/agentlens-server"]
  }
}
```

## Documentation

## What agentkitai/agentlens MCP server does

The agentkitai/agentlens MCP server is the MCP-facing component of AgentLens, an observability system for AI agents. It connects MCP clients, including Claude Desktop and Cursor, to an AgentLens instance over HTTP. The integration records MCP tool calls, sessions, and related events for later inspection through the AgentLens API and dashboard.

AgentLens is designed around an append-only audit trail. Each event is linked to the preceding event with a SHA-256 hash. Verification can identify when a record has been altered, removed, or reordered. The server also supports the broader AgentLens workflow for preserving an audit record and producing a signed JSON export through the platform's verification export endpoint.

## How it works

An MCP client sends activity through the AgentLens MCP integration. The MCP component forwards that activity to the AgentLens server, where the ingest and query components process it and store it in SQLite or Postgres. A dashboard is available from the server for viewing collected activity.

The platform can also ingest OpenTelemetry traces that use GenAI semantic conventions. OTLP HTTP traces sent to the AgentLens endpoints are mapped into events such as language-model calls, responses, tool calls, embeddings, and agent invocations. Traces become sessions, or use `gen_ai.conversation.id` when that identifier is present. These events receive the same hash-chain treatment as MCP-originated records.

## Setup and configuration

The agentkitai/agentlens MCP server is distributed as an npm package named `@agentkitai/agentlens-mcp`. The repository also documents a separate AgentLens server package, `@agentkitai/agentlens-server`, which provides the receiving service and dashboard. A local AgentLens instance can run with SQLite and zero configuration through Docker or the server package.

For a quick local deployment, the documented Docker command publishes port 3400. Production-style deployments can use the repository's Docker Compose configuration with Postgres, Redis, authentication, and TLS settings. The MCP client must be configured to connect to the relevant AgentLens deployment; the provided material does not specify a complete client configuration block or MCP-specific credential variable.

## Tools and capabilities

The agentkitai/agentlens MCP server supports:

- Recording MCP tool calls, sessions, and agent events.
- Sending MCP activity to an AgentLens HTTP service.
- Preserving events in a SHA-256 hash-chained audit log.
- Verifying whether the recorded chain remains intact.
- Working alongside OpenTelemetry GenAI trace ingestion.
- Inspecting collected activity through the AgentLens API and web dashboard.

AgentLens can reconstruct model-call cost from model identifiers and token usage for supported OpenTelemetry GenAI traces. Its documented integrations also include Python auto-instrumentation, an OpenClaw plugin, and SDKs, but those are separate from the MCP package.

## Limitations and notes

The MCP package is only one part of the deployment: it needs an AgentLens server to receive and store the activity. The material does not list the individual MCP tool names or provide a standalone storage mode for the MCP package.

The quick local Docker example disables authentication and uses a development JWT secret, so it is intended for local trials. Shared or production deployments should use authentication and a real JWT secret. OpenTelemetry authentication can be enabled for OTLP endpoints with `OTLP_AUTH_TOKEN`.

The agentkitai/agentlens MCP server is open source under the MIT license. The repository describes its audit trail as suitable for EU AI Act Article 12 record-keeping, but compliance decisions still depend on the complete deployment and operating process.

_Full upstream README: https://allmcps.com/mcp/agentkitai-agentlens/readme_