GrillMe MCP
Decision stress-testing MCP server that injects a GRILL protocol into ChatGPT Web conversations, enabling read-only interrogation of plans, state tracking, and recommended answers for decision-making.
README
GrillMe MCP
Decision stress-testing for ChatGPT Web through a remote MCP app.
Special thanks to Matt Pocock. GRILL was inspired by and evolved from Matt Pocock's excellent
/grill-me/grillingworkflow. His original idea — relentlessly interview a plan until shared understanding, resolve decision dependencies, research facts instead of asking the user, and recommend an answer for each decision — is the foundation that made this project possible.
This repository adapts that concept for normal ChatGPT Web Chat as a remotely hosted MCP app, then extends it with explicit state tracking, evidence/tool routing, contradiction handling, read-only boundaries, modern + legacy ChatGPT MCP compatibility, and a Shared Understanding approval gate.
What it does
GrillMe exposes one model-visible MCP tool:
grill
It can be invoked from a normal ChatGPT conversation after the custom app is connected, for example:
@GrillMe /grill
Quiero decidir si deberíamos rediseñar la arquitectura del producto.
Supported control commands include:
startstatustreeassumptionsfactsrisksresumestopapprove
Architecture
ChatGPT Web
↓
@GrillMe / custom app
↓
AppDeploy MCP gateway
↓
POST /api/mcp
↓
tool: grill
↓
GRILL protocol in the current conversation
The server is intentionally stateless for decision-session state: the conversation itself carries the evolving decision tree. The MCP tool injects/coordinates the GRILL protocol and remains read-only while GRILL is active.
ChatGPT registration
Current hosted diagnostic frontend:
https://grill-chat-sby65b.v2.appdeploy.ai/
Current MCP URL used to register the custom app in ChatGPT Developer Mode:
https://api-v2.appdeploy.ai/app/grill-chat-sby65b/api/mcp
Authentication: No authentication.
Repository layout
backend/index.ts— production MCP backend currently deployed through AppDeploy.index.html,src/, configs — diagnostic frontend/source.tests/tests.txt— AppDeploy QA scenarios.skill/— GRILL v2 companion skill and references..punto/— Punto por Punto Project Pack.
Original work and attribution
The original /grill-me skill is part of mattpocock/skills, created by Matt Pocock and distributed under the MIT License.
This repository is an adaptation/extension for a different runtime and distribution model. It is not presented as Matt Pocock's official ChatGPT implementation.
See THIRD_PARTY_NOTICES.md for the upstream copyright and license notice.
Governance
main is canonical. Substantive changes should land through branch → pull request → human review/merge. Project-specific canonical context lives under .punto/; the Punto por Punto mother repository can register and federate the project without duplicating its memory.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。