Settled
Enables AI agents to query a team's decision ledger to check if past decisions are still binding before taking action, via the decisions:// MCP protocol.
README
Settled — the decision layer for Slack
Retrieval tells you what was said. Settled tells you what was decided — and whether it still binds.
Teams run AI agents in Slack now. An agent acts on whatever it reads — even a decision the team already reversed. Settled maintains a decision ledger with epistemic status and, uniquely, exposes it over an MCP server so any agent can ask "is this still binding?" before it acts.
Slack Agent Builder Challenge · New Slack Agent track · Qualifying tech: our own
decisions://MCP server (ungated).
What it does
- Watches channels, detects decisions behind a precision-first confidence gate (uses thread context, catches multiple decisions per message).
- Human ✅ ratifies — only a human makes a decision binding. Wrong "settled" > silence.
- Epistemic lifecycle: proposed → contested → settled → superseded, each anchored to a verbatim quote + permalink.
- Query:
/settled <topic>, the assistant (DM/@mention), or the App Home dashboard. - MCP
decisions://—is_binding(topic),query_decisions(q)+ resources, over stdio & HTTP.
What makes it different
Plenty of tools log decisions for a person to read later. Settled is the one an agent can
query before acting — the guardrail that keeps humans and agents on what the team actually decided.
See it live: python -m demo.agent_coding_guardrail — a coding agent writes a Postgres migration,
asks Settled over MCP, sees the team ratified Aurora, and rewrites its own code before merge
(both files emitted to demo/out/). Also python -m demo.agent_guardrail for the minimal version.
Run locally
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # fill Slack tokens + OPENROUTER_API_KEY
python -m seed.seed_demo # seed the ledger (fictional demo data)
python run.py # Slack bot (Socket Mode)
python -m mcp_server.server # MCP server (stdio); SETTLED_MCP_TRANSPORT=http for HTTP
Create the Slack app from slack_manifest.yaml (declares App Home,
/settled, the Assistant surface, and events incl. reaction_removed).
Tests: pip install -r requirements-dev.txt && pytest -q — covers the ledger lifecycle,
supersede/topic guards, the hash-chained audit log (incl. tamper detection), the classifier
gates, and the MCP is_binding / verify_audit_log tools.
Deploy (always-on for judging)
See DEPLOY.md — Railway or Fly, same Dockerfile. Currently live on Railway.
Layout
| Path | Role |
|---|---|
settled/ |
config, db, ledger, llm (classify+answer), extraction, blocks, slack_app, agent |
mcp_server/server.py |
decisions:// MCP server (stdio + streamable-HTTP) |
demo/agent_guardrail.py |
live proof: external agent blocked by a superseded decision |
seed/ |
demo ledger seeders (DB-only + live-Slack) |
video/ |
Remotion video project + pipeline (the demo film) + architecture PNG |
Stack
Python · Slack Bolt · SQLite (hash-chained audit) · Model Context Protocol · OpenRouter (DeepSeek).
Licensed MIT (see LICENSE); third-party components in NOTICE.md. Demo data is fictional.
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。