Agent Check

What the agent-check MCP server does

The agent-check MCP server provides process monitoring for AI agents. Its documented scope covers three stages of a run: starting the run, recording evidence for individual tasks, and completing the run. This gives an agent workflow a way to represent progress at both the overall-run level and the per-task level.

The service is relevant when an agent performs multiple tasks and the workflow needs more than a final result. Task evidence can be submitted during execution, while run completion marks the end of the tracked process. The available material does not describe schemas, evidence formats, identifiers, or validation rules, so those details should be confirmed from the service documentation before implementation.

How it works

A workflow begins by starting a run. As work proceeds, the agent submits evidence associated with particular tasks. Once the workflow has finished, it completes the run. The documented sequence is therefore start, record task evidence, and complete.

The agent-check MCP server is intended to expose these monitoring actions to an AI agent through MCP. The supplied description does not specify whether runs can be paused, updated, queried, cancelled, or reopened. It also does not state whether the service stores results, reports status, or enforces completion conditions.

Tools and capabilities

The documented capabilities are:

• Start a monitored agent run.
• Submit evidence for a task within a run.
• Complete a monitored run.
• Track a workflow across run-level and task-level events.

These capabilities are enough to model a basic execution lifecycle, but the available material does not provide formal tool names, input parameters, return values, error behavior, or authentication requirements. Implementers should avoid assuming that evidence is free-form or that task records are automatically created.

Setup and configuration

No setup instructions are available in the supplied material. The source is listed as https://agent-check.basenull.com, but the material does not identify an installation command, package, transport, configuration file, required credential, or environment variable. It also does not name supported MCP clients.

Before adopting the agent-check MCP server, verify how the service is launched or reached, how a client registers it, and whether authentication is required. Confirm the expected run and task identifiers as well as the format accepted for evidence submissions.

Limitations and notes

The README is unavailable, so this reference is limited to the stated process-monitoring scope. No claims can be made here about integrations, persistence, dashboards, retries, access control, client compatibility, licensing, or production requirements. The agent-check MCP server should be evaluated against the workflow’s required evidence model and lifecycle behavior before use.