MCP Server

The Policy MCP server lets AI assistants evaluate, validate, and explain policy rules directly. It implements the Model Context Protocol and is built into the API — no additional installation required.

Endpoint

The MCP server is available at:

https://api.policy2.net/mcp

It uses Streamable HTTP transport — all communication happens on a single endpoint via POST /mcp.

Authentication

All MCP connections require a valid API key with view scope. Pass the key as a query parameter:

https://api.policy2.net/mcp?api_key=YOUR_API_KEY

Or use the x-api-key header on requests.

Setup

Claude Desktop

Add the following to your Claude Desktop configuration file:

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

{
  "mcpServers": {
    "policy": {
      "type": "streamable-http",
      "url": "https://api.policy2.net/mcp?api_key=YOUR_API_KEY"
    }
  }
}

Claude Code

Add the MCP server to your project with the Claude Code CLI:

claude mcp add policy --transport http "https://api.policy2.net/mcp?api_key=YOUR_API_KEY"

Or add it manually to .claude/settings.json:

{
  "mcpServers": {
    "policy": {
      "type": "http",
      "url": "https://api.policy2.net/mcp?api_key=YOUR_API_KEY"
    }
  }
}

OpenAI Codex

Add the MCP server to your codex configuration. In your project’s codex.json or via the CLI:

{
  "mcpServers": {
    "policy": {
      "type": "http",
      "url": "https://api.policy2.net/mcp?api_key=YOUR_API_KEY"
    }
  }
}

Tools

The MCP server exposes four tools that AI assistants can call.

evaluate_rules

Evaluate a policy rule against JSON data. Returns the overall result, per-outcome booleans, an execution trace, and any errors.

Parameters:

Field	Type	Description
rule	string	The policy rule DSL string to evaluate
data	object	The JSON data to evaluate against

Example call:

{
  "rule": "A **Person** gets senior_discount\n  if the __age__ of the **Person** is greater than or equal to 65.",
  "data": { "Person": { "age": 70 } }
}

Example response:

{
  "result": true,
  "outcomes": { "senior_discount": true },
  "trace": { "..." },
  "error": null
}

validate_rule

Check whether a policy rule is syntactically valid without evaluating it.

Parameters:

Field	Type	Description
rule	string	The policy rule DSL string to validate

Example response:

{
  "valid": true,
  "error": null
}

explain_rule

Parse a policy rule and return a structured breakdown of its selectors, outcomes, and conditions.

Parameters:

Field	Type	Description
rule	string	The policy rule DSL string to explain

Example call:

{
  "rule": "A **Person** gets senior_discount\n  if the __age__ of the **Person** is greater than or equal to 65."
}

Example response:

{
  "rule_count": 1,
  "rules": [
    {
      "label": null,
      "selector": "Person",
      "outcome": "senior_discount",
      "conditions": [
        {
          "type": "comparison",
          "selector": "Person",
          "property": "age",
          "operator": "is greater than or equal to",
          "value": "65",
          "logical_op": null,
          "optional": false,
          "negated": false
        }
      ]
    }
  ],
  "error": null
}

list_operators

List all comparison operators supported by the policy DSL, grouped by operator with all accepted string forms. Takes no parameters.

Example response (abbreviated):

[
  {
    "operator": "is greater than or equal to",
    "forms": ["is greater than or equal to", "is at least", ">="]
  },
  {
    "operator": "is equal to",
    "forms": ["is equal to", "is", "equals", "=="]
  },
  {
    "operator": "is a valid email",
    "forms": ["is a valid email", "is a valid email address"]
  }
]

What Can AI Assistants Do With This?

Once configured, your AI assistant can:

• Write policy rules — ask it to create rules for your business logic and validate them on the fly
• Test rules against data — provide sample JSON and have the assistant evaluate whether a rule passes or fails
• Debug rules — use explain_rule to break down a rule’s structure and evaluate_rules with trace output to see exactly which conditions pass or fail
• Discover operators — the assistant can call list_operators to find the right comparison for your use case

Example Conversation

You: Write a policy that gives free shipping to orders over $50 that weigh less than 5kg.

Assistant: (uses validate_rule to confirm syntax, then evaluate_rules to test it)

Here’s the policy:

A **Order** gets free_shipping
  if the __total__ of the **Order** is greater than 50
  and the __weight__ of the **Order** is less than 5.

I’ve validated the syntax and tested it against sample data — it works correctly.

Source

• GitHub: github.com/policies2/orchestrator — MCP is built into the API service