sui-ops-bot-mcp
MCP server for managing developer escalations: turns Slack messages into tracked issues in Google Sheets, with tools to post messages, check status, get weekly reports, mark solved, and ping.
README
Sui Dev-Leads Ops Bot
Turns a Slack channel plus one Google Sheet (the "Open Questions" tab) into an always-organized developer-escalation tracker for Sui dev-relations leads.
Two entrypoints share the same sheet as the single source of truth, so IDs are consistent across every dev lead:
| Entrypoint | What it is | What it does |
|---|---|---|
Auto-tracker (sui_ops_bot.slackbot) |
Always-on Slack Socket Mode app | Classifies new messages with an LLM, logs escalations to the sheet, tracks resolution with human-in-the-loop confirmation, serves status / open / aging reports. |
MCP server (sui_ops_bot.mcpserver) |
FastMCP tools inside Claude | Five natural-language tools for dev-leads: post_message, check_status, weekly_report, mark_solved, ping. Runs over stdio or token-gated HTTP. |
Run either, or both. See docs/mcp.md for the MCP tool guide and
docs/operations.md for deployment and day-to-day ops.
Classification taxonomy
Every escalation is classified on two independent axes (plus a priority). Both are
LLM-classified: the auto-tracker does it automatically; the MCP post_message tool
auto-classifies but lets the human override. They are written to two sheet columns
(Product, Type) the bot auto-adds; the legacy Escalated To column is left
untouched. The lists live in src/sui_ops_bot/config.py.
- product (15): DeepBook, Walrus, Harbor, Seal, Nautilus, MemWal, Enoki, Slush, zkLogin, SDK, Bridge, Sui Core, Hashi, Program, Other
- type (5): Question, Open PR, Bug, Feature Request, Communication
- priority: High, Medium, Low
Quickstart
# 1. Install (editable, with the dev toolchain)
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
# 2. Configure
cp .env.example .env # fill in the values
# plus a Google service-account json at secrets/service_account.json
# 3. Validate config (no side effects beyond token/sheet checks)
python -m sui_ops_bot.slackbot --check
python -m sui_ops_bot.mcpserver --check
# 4. Run
python -m sui_ops_bot.slackbot # auto-tracker (long-running)
python -m sui_ops_bot.mcpserver # MCP over stdio (launched by the client)
--diag runs live read-only diagnostics for the auto-tracker (Slack identity,
channel membership, one classify call, sheet access).
How it works
- Auto-log. A cheap local pre-filter drops chatter; substantive top-level
messages go to Claude, which decides if a message is a new developer escalation
and extracts the structured fields (summary, platform, product, type, priority).
If yes, a row is appended and the bot posts an in-thread note with a deep link.
React
:x:to discard a false positive. - Resolve, human-in-the-loop. A thread reply on a tracked message, or a
:white_check_mark:on it, triggers Claude to judge resolution. If it looks resolved, the bot proposes closure and asks the owner to confirm with a:white_check_mark:; the confirming reaction closes the row. Nothing is ever auto-closed. - Report.
weekly_report(MCP) groups the open backlog by product and acceptsproduct=/type=filters. The auto-tracker answers!status,!open,!agingin-channel.
The sheet holds the state: two classification columns (Product, Type) plus
three infra columns (Slack Channel, Slack TS, Bot Refs) let the in-memory
index be rebuilt on every boot, so redeploys on ephemeral hosts never lose tracking.
Layout
src/sui_ops_bot/ config, ids, logutil, classify, sheet, reports,
slack_client, slackbot (auto-tracker), mcpserver (MCP)
tests/ pytest on the pure logic (no network)
deploy/ Dockerfiles + docker-compose
docs/ MCP tool guide + operations
.claude/ SOP + git-standards + project context (CLAUDE.md at root)
Development
pytest # unit tests on the pure logic (ids, taxonomy, reports)
ruff check . # lint
The workflow (branch, TDD, review, ship) is documented in
.claude/rules/SOP.md. Task tracking is GitHub Issues only.
Security
.env and everything under secrets/ are gitignored and must never be committed.
All Slack, Anthropic, and MCP tokens plus the Google service-account json live only
in the environment or a mounted file. The remote MCP HTTP endpoint refuses to start
without MCP_HTTP_TOKEN and gates every request behind it.
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。