PaperFrame PPT Agent
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@PaperFrame PPT AgentGenerate an editable PowerPoint framework diagram from my paper method description."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
PaperFrame PPT Agent
面向科研人员的 PowerPoint 原生框架图生成与编辑 Agent。输入论文/方法描述,输出可在 PowerPoint 中继续手工编辑的原生框架图(模块、文字、箭头、分组均为原生对象),支持自然语言增量修改、人工修改保护、快照回滚和真实渲染验收。
使用场景
情景 | 推荐做法 |
论文方法章节刚写完,想要一张能放进论文、且每个模块还能继续编辑的架构图 | 1) 把方法描述整理成 |
手头有手绘草图、PDF 截图或别的工具导出的静态图,想在 PowerPoint 里逐块编辑,而不是整张当图片 | 1) 先把图做成一张 |
图基本满意,只有一个模块的名字写错了或想换个颜色,只想改这一处 | 直接在 PowerPoint 里选中该模块改文字、改填充色即可——生成的就是原生对象,秒级完成 |
想调整两个模块的位置或顺序,希望箭头跟着模块走,不断开、不穿乱 | 直接在 PowerPoint 里拖动模块,粘附到形状连接点的连接线会自动重路由 |
在 PowerPoint 里手工挪过模块、改过标题,还想让 AI 继续加一个 Loss 模块,最怕手工调整被冲掉 | 1) 先感知当前状态: |
连续改了好几版,最新一版不满意,想退回某个修改前的状态 | 1) 列出快照: |
小修小改(改字、改色、拖动)在 PowerPoint 里直接做最快;从论文出图、转静态图为原生对象、多步连续编辑则用上面的命令或 MCP(见下文「交付给消费 Agent」)让 AI 连续完成。每次写操作前系统会自动创建快照(
snapshots/),所以总能回到任意修改前版本。
Related MCP server: pptx-mcp-server
快速开始
环境搭建
独立 conda 环境,安装步骤见 SETUP.md。
手动使用
# 激活环境
conda activate paperframe
# 烟雾测试(COM 最小闭环)
python scripts\smoke_test.py# 首稿生成: 示例 Transformer 架构图
python scripts\generate_framework.py --spec fixtures\transformer_pipeline.json --task transformer --out-dir projects\transformer其中
--spec: framework_spec JSON 文件路径(结构规格,见fixtures/示例)--task: 任务 id,作为输出文件名前缀(默认取 spec.project_id)--out-dir: 输出目录(生成<task>.pptx/<task>_preview.png/<task>.spec.json/<task>.run.json)
# 增量编辑: 移动/改字(需先生成 edit_plan,见 fixtures\edit_move_decoder.json)
python scripts\edit_framework.py --pptx projects\transformer\transformer.pptx --plan fixtures\edit_move_decoder.json `
--task transformer-edit --out-dir projects\transformer-edit --spec projects\transformer\transformer.spec.json其中
--pptx: 目标 .pptx 绝对路径--plan: edit_plan JSON 文件路径(操作计划,见fixtures/示例)--task: 任务 id,用于快照与输出命名--out-dir: 新版本输出目录(保存为 vNNN,不覆盖原文件)--spec: 生成时的 spec 文件(可选,供结构 QA 对比)--confirm: 高风险操作(删除节点/整页重建/覆盖原文件)需加上此参数确认
# 回滚
python scripts\edit_framework.py --rollback <snapshot_id> --task transformer-edit --out-dir projects\transformer-edit\restored其中
--rollback: snapshot_id(可用--task对应的快照列表查看,见下)--task: 快照所属任务--out-dir: 恢复目标目录
查看任务快照:
python -m paperframe.cli snapshots --task transformer-edit无 PowerPoint 的环境用
--no-export跳过预览导出与视觉 QA。
Agent自动化(MCP + Skill)
本项目可交付给agent 直接使用: LLM 做语义规划(产出 framework_spec.json / edit_plan.json),MCP server 做确定性执行
conda activate paperframe
python -m pip install -e . # 安装包 + paperframe / paperframe-mcp 命令
python -m paperframe.mcp_server # 启动 MCP server(stdio,默认)
# 远端: python -m paperframe.mcp_server --transport streamable-http --host 127.0.0.1 --port 8765工具(9 个):
generate: 从 spec 生成原生 PPTX + QAedit: 增量编辑(快照→原子操作→新版本 vNNN)inspect: 感知人工改动 / 同步当前状态回 specvalidate: 结构 QA(校验 PPTX 与 spec 一致性)import_pptx: 反向导入任意 PPTX 为 specvalidate_spec: 校验 framework_spec JSON(语义规划后自检)validate_plan: 校验 edit_plan JSON(编辑计划自检)rollback: 从快照恢复文件list_snapshots: 列出任务的全部快照
资源(8 个):
paperframe://schema/framework_spec: framework_spec JSON Schemapaperframe://schema/edit_plan: edit_plan JSON Schemapaperframe://design_tokens: 默认样式令牌(字体/配色/线型)paperframe://prompt/planner: 语义规划 prompt(论文→spec)paperframe://prompt/edit_plan: 增量编辑 prompt(自然语言→edit_plan)paperframe://skill: PaperFrame 使用总览(触发/路由/门槛/交付物)paperframe://templates/components: framework_spec 组件模板库(节点/边/分组写法)paperframe://templates/edit_ops: edit_plan 操作模板库(22 个 op 写法)
消费 agent 的触发/路由/门槛说明见 skill/paperframe-ppt/SKILL.md
约定
对象命名:
node_<semantic_id>/edge_<from>_<to>/group_<semantic_id>/label_<semantic_id>所有写操作前创建快照(
snapshots/),失败可回滚交付前必须真实 PowerPoint 渲染 + 视觉 QA;阻断性问题不得交付
高风险操作(整页重建、删除节点、覆盖原文件等)必须向用户确认
增量编辑默认保留人工修改;连接线优先粘附到形状连接点,移动节点后跟随重路由
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Flicense-qualityAmaintenanceMCP server for generating slides from natural language, with interactive preview and export to PDF, PPTX, and Markdown.18
- Alicense-qualityBmaintenanceMCP server for creating, reading, and editing PowerPoint (.pptx) presentations.1MIT
- Flicense-qualityBmaintenanceMCP server for AI-native editable presentations: surgically edit slide components by stable ID, with local rendering and self-checking overflow.
- Alicense-qualityCmaintenanceCross-platform MCP server for Microsoft PowerPoint automation, enabling slide management, template remapping, and real-time evaluation via natural language.MIT
Related MCP Connectors
MCP server for generating rough-draft project plans from natural-language prompts.
Presentations.AI MCP server — create designed slide decks from a topic, text, or document.
MCP server for the Fail Modes taxonomy — a knowledge base of AI system failure modes
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/MyLiya46/paperfream'
If you have feedback or need assistance with the MCP directory API, please join our Discord server