Erdl Mcp

Deterministic rules for AI Agents. Say it once. Never repeat.

npm version npm downloads license tests

📖 中文版

ERDL (Entity-Rule Definition Language) gives your Agent deterministic rules. 30 built-in rules across 6 categories. Unlimited personal rules. Free forever.

---

Why ERDL?

Without ERDL	With ERDL
"Don't use any" — forgotten in 5 turns	Rules enforced deterministically, every tool call
"Write shorter sentences" — ignored	Writing rules fire before every output
"Confirm before adding deps" — skipped	Blocked until you confirm
"Why did you do that?" — no answer	erdl_explain shows the full decision trail

---

Quick Start

# Start MCP Server (5 tools, 30 rules — all MCP clients)
npx -y @openoba-ai/erdl-mcp

# Or: install OpenClaw Plugin for hardware enforcement
openclaw plugins install @openoba-ai/erdl-openclaw

30 rules active immediately. No account. No configuration. No API key.

中文用户：

npx -y @openoba-ai/erdl-mcp --lang zh

---

Free vs Pro

	Free	Pro
5 MCP Tools	✅	✅
30 Preset Rules	✅	✅
Unlimited Personal Rules	✅	✅
Chinese / English	✅	✅
All 11 Operators	✅	✅
MIT Open Source	✅	✅
Execution Rings 1–2 (REQUEST_HUMAN, ESCALATE, ROLLBACK, QUARANTINE)	—	✅
Guardian Agent Role (监管者模式)	—	✅
Audit Export (OCSF / OTLP)	—	✅
Team Rules (组织级规则共享)	—	✅
Dashboard (规则命中统计)	—	✅
Enterprise Compliance Packs (GB/Z 185, NIST, EU AI Act)	—	Enterprise

Get a Pro License →

One-command Install

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/OpenOBA/erdl-mcp-server/master/scripts/install.sh | bash

# Windows (PowerShell)
iwr https://raw.githubusercontent.com/OpenOBA/erdl-mcp-server/master/scripts/install.ps1 | iex

---

Why ERDL vs SKILL.md / Prompt Rules?

	Prompt-Based Rules	ERDL MCP Server
Execution	LLM "tries" to follow	Deterministic engine — guaranteed
Visibility	Invisible, can't tell if it worked	erdl_explain shows every decision
Reliability	LLM may ignore or forget	Zero hallucination — condition match is mathematical
Testing	Manual	erdl_simulate with 3 auto-scenarios
Portability	Single platform	All MCP-compatible Agents

ERDL rules don't suggest. They don't hope. They enforce.

---

What's Included

30 Built-in Rules (8 Categories)

Category	Rules	Scope
coding	10	TypeScript quality, Git discipline, dependency hygiene
engineering	24	Workflow discipline, pipeline gates, session memory, self-verify
writing	7	Tone, formatting, clarity
design	3	Tailwind-first, responsive, accessible
security	6	SQL injection, eval, hardcoded secrets, input validation
testing	11	Coverage, boundary conditions, flaky tests, TDD workflow
performance	3	N+1 queries, pagination, measure before optimize
observability	3	Structured logging, health checks, no secrets in logs

5 MCP Tools

Tool	Purpose
erdl_evaluate	Action Guard — check rules before every tool call
erdl_simulate	Test a rule against 3 scenarios before creating
erdl_create_rule	Create a rule from natural language
erdl_list_rules	List all active rules
erdl_explain	Full decision trail for any action

MCP Resources

Resource	Content
erdl://rules/list	All rules as JSON
erdl://status	Server runtime status

---

Installation

Any MCP Client

{
  "mcpServers": {
    "erdl": {
      "command": "npx",
      "args": ["-y", "@openoba-ai/erdl-mcp@latest"]
    }
  }
}

Or let ERDL generate the config for your client:

npx @openoba-ai/erdl-mcp@latest --setup

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent path on your OS.

Cursor

Add to .cursor/mcp.json in your project root.

OpenClaw

# MCP Server (all MCP clients)
openclaw mcp set erdl '{"command":"npx","args":["-y","@openoba-ai/erdl-mcp@latest"]}'

# OpenClaw Plugin (hard enforcement via before_tool_call)
openclaw plugins install @openoba-ai/erdl-openclaw

Plugin vs MCP Server: The MCP Server gives your Agent ERDL tools. The Plugin adds deterministic tool call interception — rules run BEFORE every exec/write/search, no relying on LLM self-discipline. Install both for maximum protection.

VS Code / GitHub Copilot

Add to .vscode/mcp.json or the Copilot MCP configuration.

---

Creating Your First Rule

You: "Never use `any` in TypeScript."

Agent: 1. erdl_simulate → tests the rule against 3 scenarios
       2. erdl_create_rule → saves to ~/.openoba/rules/
       3. Rule is now active. Next `any` will be blocked.

You: "Why did you reject my code?"
Agent: erdl_explain → shows every rule that fired

Or create rules manually in ~/.openoba/rules/ (ERDL SPEC §5 format):

protocol: "erdl/v1"
version: "1.0.0"

rules:
  - name: "no_console_log"
    description: "Don't commit console.log statements"
    priority: 10
    when:
      logic: AND
      conditions:
        - field: "tool.name"
          operator: in
          value: ["write", "edit", "apply_patch"]
        - field: "tool.args"
          operator: match
          value: "console\\.log"
    then: DENY
    message: "Remove console.log before committing"

📚 Full tutorial: Create Your First ERDL Rule → (English)

📖 Complete Rule Reference (all 30 rules with modification guides) → (English)

---

CLI Reference

npx @openoba-ai/erdl-mcp@latest          # Start MCP server (always latest)
npx @openoba-ai/erdl-mcp@latest --lang zh # Chinese mode
npx @openoba-ai/erdl-mcp@latest --pro-key sk-xxx  # Activate Pro license
npx @openoba-ai/erdl-mcp@latest --upgrade  # Upgrade to latest
npx @openoba-ai/erdl-mcp@latest --uninstall  # Clean removal
npx @openoba-ai/erdl-mcp@latest --help     # Full usage

Upgrading from v1.0.x? You MUST update your MCP config:

Remove old local-path config:

openclaw mcp remove erdl  # or remove "erdl" from mcp.servers in config

Re-add with npx (auto-latest):

openclaw mcp set erdl "{\"command\":\"npx\",\"args\":[\"-y\",\"@openoba-ai/erdl-mcp@latest\"]}"

Or manually in ~/.openclaw/openclaw.json → mcp.servers.erdl:

{ "command": "npx", "args": ["-y", "@openoba-ai/erdl-mcp@latest"] }

npm cache clearing (if --version shows an older release):

npm cache clean --force

Run from Source

git clone https://github.com/OpenOBA/erdl-mcp-server.git
cd erdl-mcp-server
npm install
npm run build
node bin/erdl-mcp.js --version

---

License

MIT · OpenOBA · @OpenOBA

Built by AI + Human. Deterministic architecture, not prompt engineering.