What can you do with this server?

The ERDL MCP Server provides deterministic rule enforcement for AI agents, ensuring consistent, compliant behavior through mathematical condition matching rather than relying on LLM prompts. * Evaluate/Guard tool calls (erdl_evaluate): Intercept and check any planned tool call against loaded rules before execution, returning decisions like ALLOW, DENY, CORRECT, REQUEST_HUMAN, or EMERGENCY_HALT. * List active rules (erdl_list_rules): View all currently loaded rules, optionally filtered by category (coding, engineering, writing, design, security, performance, testing, compliance, accessibility, or custom). * Create rules from natural language (erdl_create_rule): Turn plain-language instructions (e.g., "Never use any in TypeScript") into persistent YAML rules saved to ~/.openoba/rules/, taking effect immediately. * Simulate rules before creating them (erdl_simulate): Test a proposed rule against 3 auto-generated scenarios to validate it works as intended before committing. * Explain decisions (erdl_explain): Get a full decision trail showing every rule checked and why a particular decision was made. * Access rules as resources: Query erdl://rules/list for all rules as JSON, or erdl://status for the server's runtime status. * Use 30 built-in preset rules across coding, engineering, writing, and design categories — active immediately with no configuration. * Integrate with any MCP-compatible client: Works with Claude Desktop, Cursor, OpenClaw, VS Code/GitHub Copilot, and more. * Pro/Enterprise features (with license): Includes advanced execution rings (ESCALATE, ROLLBACK, QUARANTINE), Guardian Agent Role, Audit Export (OCSF/OTLP), Team Rules, a Dashboard, and compliance packs for standards like NIST and EU AI Act.

Which integrations are available for this server?

Integrates with GitHub Copilot via VS Code MCP configuration, allowing Copilot to follow enforced rules.

How do I use ERDL MCP Server?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@ERDL MCP Server Create a rule that bans console.log in TypeScript files" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

ERDL MCP Server

by OpenOBA

Overview Schema Source Related Servers Discussions Admin

TypeScript

Local

ERDL MCP Server

Stop reminding your Agent to behave. Teach it. One when/then sentence is all it takes.

📖 中文版

npx -y @openoba-ai/erdl-mcp

30 rules. 5 tools. Zero config. Unlimited custom rules. Free forever.

ERDL demo — intercepting rm -rf /

The Problem

You tell your Agent "don't use any", "keep it short", "ask before adding dependencies". It nods. Five turns later, it's back to its old habits.

Prompt-based rules don't work. LLMs forget. They reinterpret. They rationalize. You're not guiding them — you're negotiating with them.

Related MCP server: Claude Critical Rules MCP Server

The Solution

ERDL (Entity-Rule Definition Language) is a deterministic rule engine that runs as an MCP Server. Your Agent doesn't try to follow rules — the engine enforces them before every tool call.

You say	What actually happens
"Never use `any`"	Blocked. The engine intercepts `write_file` calls, scans the content, and rejects `any` before the write happens.
"Don't start with 'In today's world'"	Denied. Writing rules fire before every output. The Agent receives the correction before it speaks.
"Ask before adding npm dependencies"	Intercepted. `exec` of `npm install` is stopped. You approve it, or it doesn't happen.
"Why did you do that?"	Explained. `erdl_explain` shows exactly which rule fired, what condition matched, and what happened.

How It Works

┌─────────────────────────────────────────────────────────┐
│                    Agent Workflow                        │
│                                                         │
│  User says: "Write me a function"                       │
│       │                                                 │
│       ▼                                                 │
│  LLM plans: use write_file(path, code)                  │
│       │                                                 │
│       ▼                                                 │
│  ┌──────────────────────────────────────┐               │
│  │    ERDL Action Guard (MCP Server)    │  ← Protocol   │
│  │                                      │    Layer      │
│  │  1. Load 30 built-in rules           │               │
│  │  2. Evaluate: field/operator/value   │               │
│  │  3. Decision: ALLOW / DENY / CORRECT │               │
│  │  4. Return badge card + explanation  │               │
│  └──────────────────────────────────────┘               │
│       │                                                 │
│       ▼                                                 │
│  Tool executes (or is blocked)                          │
│                                                         │
│  No prompt engineering. No negotiation. Protocol-level. │
└─────────────────────────────────────────────────────────┘

Quick Start

# One command. 30 rules active immediately.
npx -y @openoba-ai/erdl-mcp

That's it. No account. No API key. No configuration.

Chinese / 中文：

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

Add it to your MCP client:

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

Or let ERDL generate the config:

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

Supported clients: Claude Desktop · Cursor · VS Code / Copilot · OpenClaw · WorkBuddy · any MCP-compatible client.

ERDL vs Prompt Rules

	Prompt / SKILL.md	ERDL
Enforcement	LLM "tries" to follow	Deterministic engine — guarantees execution
Reliability	May forget after 5 turns	Condition match is mathematical — zero hallucination
Visibility	Can't tell if it worked	`erdl_explain` shows every decision, every rule
Testing	Manual verification	`erdl_simulate` — 3 auto-generated scenarios
Portability	Tied to one platform	All MCP-compatible Agents
Overridability	LLM can "reinterpret"	Protocol-layer block — Agent cannot bypass

What's Included

30 Built-in Rules

Category	Count	Covers
`engineering`	13	Pipeline gates, no shortcuts, self-verify, decision logging, no force-push, no stash
`coding`	6	No `any`, no `@ts-ignore`, naming conventions, one-commit-one-change, dependency hygiene
`security`	6	No eval with input, no hardcoded secrets, no string SQL, validate all input, security headers, no stack traces
`testing`	2	Coverage never drops, no behavior without test
`writing`	2	Direct tone, no AI jargon
`observability`	1	No secrets in logs

5 MCP Tools

Tool	Use it when
`erdl_evaluate`	Before every tool call — mandatory Action Guard
`erdl_simulate`	Test a rule against 3 scenarios before creating it
`erdl_create_rule`	User says "remember this" → create a rule from NL
`erdl_list_rules`	User asks "what rules are active?"
`erdl_explain`	User asks "why did you do that?" → full decision trail

2 Resources

Resource	Purpose
`erdl://rules/list`	All active rules as JSON
`erdl://status`	Runtime status: rule counts by category, agent role

Creating a Rule

Turn your pain points into rules, one sentence at a time — banned, period. Turn your experience into rules, one sentence at a time — shared, forever.

Every pain point is a when(…) then(…) sentence:

when [trigger] · then [action]

Variable	Meaning	Example
when	What triggers it	code contains `any` · logs contain secrets · `git stash` used
then	What happens	Block it · Correct it · Allow with a warning · Pause for approval

Fill in the blanks:

when code contains any then block it.

That's a rule.

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

Agent → 1. erdl_simulate   → tests against 3 scenarios
        2. erdl_create_rule → saves to ~/.openoba/rules/
        3. Rule is active.  → next `any` is blocked.

You →   "Why was my code rejected?"

Agent → erdl_explain → full decision trail:
         ✅ no_any: tool.name in [write_file, edit, apply_patch] AND content contains "any"
         → DENY: Do not use `any` type.

Or write rules by hand 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 → · 📖 Rule Reference (30 rules) →

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	—	✅
Audit Export (OCSF / OTLP)	—	✅
Team Rules	—	✅
Dashboard (hit statistics)	—	✅
Enterprise Compliance (GB/Z 185, NIST, EU AI Act)	—	Enterprise

Get a Pro License →

CLI Reference

npx @openoba-ai/erdl-mcp@latest           # Start (auto-latest via npx)
npx @openoba-ai/erdl-mcp@latest --lang zh  # Chinese mode
npx @openoba-ai/erdl-mcp@latest --upgrade   # Force upgrade
npx @openoba-ai/erdl-mcp@latest --uninstall # Clean removal
npx @openoba-ai/erdl-mcp@latest --setup     # Show MCP config
npx @openoba-ai/erdl-mcp@latest --help      # Full usage

From Source

git clone https://github.com/OpenOBA/erdl-mcp-server.git
cd erdl-mcp-server
npm install
npm run build
npm test        # 92 tests
node bin/erdl-mcp.js

Contributing

ERDL MCP Server is MIT-licensed and open to contributions. See CONTRIBUTING.md for guidelines.

Star History

If ERDL helps your workflow, give us a ⭐ on GitHub — it helps others discover the project.

License

MIT · OpenOBA · @OpenOBA

Deterministic architecture, not prompt engineering.

Available Tools

5 tools

erdl_create_ruleCreate ERDL RuleA

Create a new ERDL rule from natural language. Use this when the user corrects your behavior and wants you to "remember" it.

EXAMPLE SCENARIOS:

User: "Never use 'any' types" → Create: coding rule, intent: "write typescript code", DENY if "any" appears
User: "Don't start with 'in today's world'" → Create: writing rule, intent: "write blog post", DENY
User: "Always use Tailwind, never inline styles" → Create: design rule, intent: "create UI", ALLOW with instruction

The rule is saved to ~/.openoba/rules/ and takes effect immediately (no restart needed).

Inputs: category (required), decision (required), instruction (required), naturalLanguage (required), +2 more.

erdl_evaluateERDL Action Guard — Tool Call InterceptorA

ERDL Action Guard — evaluates planned tool call against loaded rules BEFORE execution.

YOU MUST call this BEFORE every tool call (exec, write_file, edit, web_search, etc). This is MANDATORY. The Guard cannot be bypassed.

After calling, check the decision:

ALLOW → proceed with the tool call, follow any instructions
DENY → STOP. Do not execute the tool. Tell the user what was blocked.
CORRECT → fix the request and retry
REQUEST_HUMAN → ask the user for approval before proceeding

Always show the result to the user as a badge card.

RESPONSE (compact badge card format):

"ALLOW": ✅ ERDL Guard · N rules
"DENY": 🛑 ERDL Blocked · reason
"REQUEST_HUMAN": 👤 ERDL Approval · reason
"CORRECT": 🔧 ERDL Correct · correction
"EMERGENCY_HALT": 🚨 ERDL HALT · reason
"PASS": 🔵 ERDL Pass · N rules checked

Inputs: tool_name (required), agent_id, tool_args, session_id.

erdl_explainExplain ERDL DecisionA

Show the FULL decision trail for the last action. Answers "why did you do that?" — shows every rule that was checked and whether it fired.

Use this when:

User asks "why did you act that way?"
User is confused about a DENY or unexpected ALLOW
You want to show transparency in your decision-making

Inputs: tool_name (required), tool_args.

erdl_list_rulesList ERDL RulesA

List all currently loaded ERDL rules. Use this when the user asks "what rules do you have?" or wants to see what constraints are active. Categorize by type: coding, engineering, writing, design, security, performance, testing, compliance, accessibility, custom.

Inputs: category.

erdl_simulateSimulate ERDL RuleA

Test a potential rule against 3 scenarios BEFORE creating it. This prevents "wishful thinking" rules that sound right but don't work.

Always call this BEFORE erdl_create_rule when the user says "remember this" or "create a rule". Show the simulation results and ask if the user wants to proceed.

Inputs: category (required), decision (required), keywords (required), ruleName (required), +2 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

Deploy Server

Claude Code, inside the Godot 4 editorFlockbay puts an AI agent in the real Godot 4 editor: it writes GDScript, runs your game and fixes what breaks. Free, or bring your own Claude Code or Codex.flockbay.comAd

TDQS

A4.3/5.0

Scored 2026-07-10 08:30 across 5 tools

Disambiguation5/5

Each tool has a clearly distinct purpose: create, evaluate, explain, list, simulate. No overlap or ambiguity between them.

Naming Consistency5/5

All tools use a consistent 'erdl_' prefix and verb-based naming (e.g., create_rule, evaluate, explain, list_rules, simulate). Minor variation in noun inclusion is acceptable.

Tool Count5/5

5 tools is well-scoped for a rule management server. Each tool covers a necessary function without redundancy or bloat.

Completeness3/5

The tool set covers creation, evaluation, explanation, listing, and simulation, but lacks update and delete capabilities for rules, which are notable gaps for a rule management system.

Resources

Related MCP Connectors

Agent ModuleA
dev.agent-module
Deterministic compliance and vertical knowledge bases for autonomous agents. Free 24hr trial.
Autonomous Agents Knowledge & Memory RAG Systems
722
https://api.agent-module.dev/mcp
boundaryai
ai.boundaryai
Deterministic runtime safety for AI agents: scan PII, gate tool actions, verify LLM output.
Security AI & Machine Learning Agent Orchestration
2
https://mcp.boundaryai.ai/mcp
THE RETREATA
com.agentsretreat
Rules-based context reduction, planning checks, research prompts and fixed experiences for agents.
Agent Orchestration Autonomous Agents
251
https://agentsretreat.com/mcp
PolicyLayer
com.policylayer
The system of record for AI agent authority: playbooks, routed policy questions, reusable rules.
Autonomous Agents
2
https://mcp.policylayer.com/mcp

Related MCP Servers

focal-mcp
Developer Tools Coding Agents
BAESI
A
license
Not graded
quality
D
maintenance
Centralizes AI behavior rules and applies them across tools like Codex, Claude Code, and Cursor, enabling agents to fetch up-to-date rules before responding.
Updated 9 months ago (2026-02-04 09:58 UTC)
1
MIT
Claude Critical Rules MCP
Developer Tools Documentation Access Testing & QA Tools
optimaquantum
A
license
A
quality
D
maintenance
Enforces 21 critical rules derived from 96+ documented failure patterns to prevent common AI assistant mistakes. Provides automatic compliance checklists, rule summaries, and auto-updates.
Updated 10 months ago (2026-01-10 13:29 UTC)
5
16 npm
7
MIT
gnt MCP Serverofficial
Legal & Compliance Autonomous Agents
gnt-ai
A
license
Not graded
quality
B
maintenance
Enables AI agents to query live, human-approved rules before taking actions, ensuring compliance and reducing errors.
Updated a day ago (2026-11-14 06:25 UTC)
30
Apache 2.0
signet-eval
Security Developer Tools
wandercom
A
license
Not graded
quality
B
maintenance
Deterministic policy enforcement for AI agent tool calls. It evaluates every tool call against user-defined rules before execution, with no LLM in the authorization path.
Updated 9 days ago (2026-11-06 15:48 UTC)
4
MIT

ERDL MCP Server

The Problem

The Solution

How It Works

Quick Start

ERDL vs Prompt Rules

What's Included

30 Built-in Rules

5 MCP Tools

2 Resources

Creating a Rule

Free vs Pro

CLI Reference

From Source

Contributing

Star History

License

Available Tools

Tool Schema Changelog

TDQS

Resources

Related MCP Connectors

Related MCP Servers

focal-mcp

Claude Critical Rules MCP

gnt MCP Serverofficial

signet-eval