Agent Evidence MCP
A local evidence recorder that captures screenshots, recordings, notes, and summaries during long-running agent tasks, enabling reviewable task records with timelines and artifacts.
README
Agent Evidence MCP
中文
让 agent 在执行长任务时,自动留下截图、短录屏、备注和最终总结。
Agent Evidence MCP 不是普通截图软件,也不是浏览器自动化框架。它更像一个本地证据记录器:当 agent 正在操作浏览器、桌面软件、后台系统或多步骤流程时,它把关键状态保存成可复盘的任务记录。
适合什么场景
- agent 正在执行多步骤 UI、QA、运维、配置或排障任务
- 你希望关键节点自动留图,失败或异常时额外留证据
- 任务结束后需要
summary.md、timeline.jsonl和所有关键产物路径 - 你想让同事、客户或之后的自己快速复盘 agent 做过什么
装上之后能做什么
- 创建任务会话:
start_session - 查看已有会话:
list_sessions/get_session/get_latest_session - 保存里程碑截图:
capture_checkpoint - 必要时录一小段屏幕:
start_recording/stop_recording - 给证据加备注或 OCR 文本:
attach_note/ocr_artifact - 生成可分享的打码副本:
redact_artifact - 比较最近的证据变化:
compare_artifacts/compare_latest_artifacts - 结束时生成可交付的任务记录:
end_session - 通过 MCP resources 读取最新 summary、artifact 列表和 session 索引
- 通过 MCP prompts 获取推荐的证据采集和最终复盘提示
推荐用法
这个项目最适合配合 agent 使用。通常只需要这样告诉 agent:
Use agent-evidence MCP for this task.
Start a session first.
Capture a screenshot at each major milestone and an extra one on errors.
Prefer screenshots over recording unless motion matters.
When the task is done, give me the summary and the key artifact paths.
3 步上手
- 安装
从 GitHub 安装当前版本:
pip install git+https://github.com/xiexie-qiuligao/agent-evidence-mcp.git
如果你安装的是已发布到 Python 包索引的版本:
pip install agent-evidence-mcp
如果你是开发者,已经 clone 了这个仓库:
pip install -e .
- 初始化配置
agent-evidence-mcp init
- 启动 MCP 服务
agent-evidence-mcp serve
命令行快速体验
创建一个 session:
agent-evidence-mcp start-session "Admin QA Flow"
截一张关键节点图:
agent-evidence-mcp capture-checkpoint "D:\path\to\session" "form-submitted" "The form was submitted successfully."
结束并生成总结:
agent-evidence-mcp end-session "D:\path\to\session"
输出结构
artifacts/
<session_id>/
session.json
timeline.jsonl
summary.md
details/
screenshots/
recordings/
最终拿到的不是一堆散乱截图,而是一套带时间线、摘要和元数据的任务证据包。
平台支持
| 能力 | Windows | macOS | Linux |
|---|---|---|---|
| Session / MCP / CLI | 已实现并本地验证 | 已实现 | 已实现 |
| 截图 | 已实现并本地验证 | 已实现,尚未在本仓库实机验证 | 已实现,依赖本地截图工具 |
| 录屏 | 已实现并本地验证 | 已实现,尚未在本仓库实机验证 | 已实现,依赖 X11/ffmpeg 环境 |
| Redaction | 已实现并本地验证 | 暂未实现 | 暂未实现 |
更详细的说明见 support-matrix.md。
校验下载文件
release 页面会附带 SHA256SUMS.txt。如果你下载了 wheel 或源码包,可以这样校验:
certutil -hashfile dist\\agent_evidence_mcp-0.1.0a1-py3-none-any.whl SHA256
certutil -hashfile dist\\agent_evidence_mcp-0.1.0a1.tar.gz SHA256
文档
English
Let your agent leave behind screenshots, short recordings, notes, and a final summary while it works through a long-running task.
Agent Evidence MCP is not a generic screenshot app or a browser automation framework. It is a local evidence recorder for agent work: when an agent operates a browser, desktop app, admin console, or multi-step workflow, this server saves the important states as a reviewable task record.
Good Fits
- multi-step UI, QA, ops, configuration, or troubleshooting tasks
- workflows where milestone screenshots and error evidence matter
- handoffs that need
summary.md,timeline.jsonl, and artifact paths - reviews where another person needs to understand what the agent did
What You Get
- start a task session with
start_session - inspect existing sessions with
list_sessions,get_session, andget_latest_session - capture milestone screenshots with
capture_checkpoint - record short screen segments when motion matters
- attach notes or OCR text to artifacts
- generate redacted copies for safer sharing
- compare recent artifacts for review-oriented changes
- finish with
summary.md,timeline.jsonl, and organized artifacts - read the latest summary, artifact list, and session index through MCP resources
- use MCP prompts for evidence capture and final review guidance
Best Way To Use It
This project is designed to work with an agent. In most cases you can simply tell the agent:
Use agent-evidence MCP for this task.
Start a session first.
Capture a screenshot at each major milestone and an extra one on errors.
Prefer screenshots over recording unless motion matters.
When the task is done, give me the summary and the key artifact paths.
Quick Start In 3 Steps
- Install
Install the current GitHub version:
pip install git+https://github.com/xiexie-qiuligao/agent-evidence-mcp.git
If you are installing a published Python package:
pip install agent-evidence-mcp
If you are developing from a local clone:
pip install -e .
- Initialize config
agent-evidence-mcp init
- Start the MCP server
agent-evidence-mcp serve
Try It From The CLI
Create a session:
agent-evidence-mcp start-session "Admin QA Flow"
Capture a milestone:
agent-evidence-mcp capture-checkpoint "D:\path\to\session" "form-submitted" "The form was submitted successfully."
End the session:
agent-evidence-mcp end-session "D:\path\to\session"
What The Output Looks Like
artifacts/
<session_id>/
session.json
timeline.jsonl
summary.md
details/
screenshots/
recordings/
The result is not a pile of loose screenshots. It is a reviewable task evidence package with a timeline, summary, and artifact metadata.
Platform Support
| Capability | Windows | macOS | Linux |
|---|---|---|---|
| Session / MCP / CLI | Implemented and locally validated | Implemented | Implemented |
| Screenshots | Implemented and locally validated | Implemented, not locally validated in this repo | Implemented, depends on local screenshot tools |
| Recording | Implemented and locally validated | Implemented, not locally validated in this repo | Implemented, depends on X11/ffmpeg setup |
| Redaction | Implemented and locally validated | Not implemented | Not implemented |
See support-matrix.md for more detail.
Verify Downloads
The release includes SHA256SUMS.txt. If you download the wheel or source archive, you can verify them like this:
certutil -hashfile dist\\agent_evidence_mcp-0.1.0a1-py3-none-any.whl SHA256
certutil -hashfile dist\\agent_evidence_mcp-0.1.0a1.tar.gz SHA256
Docs
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.