Precedent 📌 https://precedent-website.vercel.app/

Decision memory for Slack. Watches team channels, extracts decisions with Claude, stores them with permalinks, answers "what did we decide about X?" in the assistant panel with receipts, nudges action-item owners, and exposes the decision log to external agents via MCP.

Built for the Slack Agent Builder Challenge. Uses all three required technologies: Slack AI assistant surface, the Real-Time Search API (assistant.search.context), and an MCP server.

Setup

1. Create the Slack app from manifest.json at api.slack.com/apps (Create New App → From a manifest), enable the Agents/AI toggle, enable Socket Mode, generate an app-level token with connections:write, and install to the workspace.
2. cp .env.example .env and fill in every value (see comments in the file).
3. npm install

Run

npm start          # the app: Bolt (Socket Mode) + REST bridge on :3000
npm run seed       # post demo conversations to 4 channels (prints channel IDs)
npm run mcp        # stdio MCP server (Claude Desktop launches this via config)

After seeding, sweep each channel through the extractor (oldest threads first is handled automatically):

curl -X POST "http://localhost:3000/api/dev/scan?channel=CXXXX" -H "Authorization: Bearer $MCP_API_TOKEN"
curl -X POST "http://localhost:3000/api/dev/nudge"              -H "Authorization: Bearer $MCP_API_TOKEN"
curl "http://localhost:3000/health"

Smoke test: @Precedent which email vendor? in any seeded channel — it must reply with Real-Time Search matches and permalinks.

Claude Desktop (MCP)

%APPDATA%\Claude\claude_desktop_config.json (Windows) / ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "precedent": {
      "command": "node",
      "args": ["C:\\Users\\danis\\OneDrive\\Desktop\\Slack\\mcp\\server.js"],
      "env": {
        "BRIDGE_URL": "https://<your-railway-domain>.up.railway.app",
        "MCP_API_TOKEN": "<same value as the deployed bridge>"
      }
    }
  }
}

Fully restart Claude Desktop after editing. Tools: search_decisions, get_decision, list_open_action_items. Every call is logged to stderr ([MCP] search_decisions("…")); Claude Desktop captures it in %APPDATA%\Claude\logs\mcp-server-precedent.log.

Deploy your own

Precedent is MIT-licensed — run it for your own team:

1. Create a Slack app in your workspace from manifest.json (Create New App → From a manifest). Enable the Agents/AI toggle, generate an app-level token with connections:write, install to the workspace.
2. Copy .env.example → .env and fill in your tokens, your Anthropic API key, and your Slack member ID for nudges.
3. npm install && npm start — that's a fully working local instance.
4. For always-on hosting, follow the Railway steps below (any Node host with a persistent disk works).

Skip npm run seed if you want it watching real conversations only — it exists to create demo data.

Deploy (Railway)

• Node service from this repo (Railpack default builder; engines.node pins 22.x).
• Volume mounted at /data; set DB_PATH=/data/precedent.db.
• Env vars: paste .env contents in the Variables Raw Editor.
• Settings → Networking → Generate Domain (for /health + the MCP bridge).
• Leave the "Serverless" toggle OFF — the app must stay up 24/7.
• Never run npm start locally while the deployed instance is live: two Socket Mode connections split events.