MCP Server

The Rule Repository exposes a Model Context Protocol (MCP) server that allows AI agents to search, evaluate, govern, and retrieve rules directly through the MCP interface.

Transport Modes

Mode	Use Case	URL
stdio	Local agents (e.g., Claude Code running locally)	Launched as a subprocess
streamable-http	Remote agents, multi-agent setups	`http://localhost:8001/mcp`

Tools

The MCP server provides 24 tools organized into four categories: search and context, evaluation and compliance, governance, and cross-organizational.

Search and Context

`search_rules`

Search the rule corpus by keyword or natural language query.

Parameters: query (string), scope (string, optional), limit (integer, optional)

`explain_rule`

Get a detailed explanation of a rule, including its rationale, context, preconditions, exceptions, and relationships.

Parameters: rule_id (string)

`get_rules_for_context`

Retrieve rules applicable to specific files, formatted for agent consumption. This is the primary tool for agent integrations.

Parameters: file_paths (array of strings), format (string, optional), federation (string, optional)

Format options:

Format	Description	Token Budget
`instructions`	Concise natural-language directives	~500 tokens
`checklist`	Markdown checkboxes for each rule	Medium
`detailed`	Full rule metadata including scope, modality, tags, and source	Largest

Default format is instructions.

When federation is provided (a federation node ID), rules are resolved through the federation hierarchy for that node, returning the effective rule set after applying inheritance and overrides.

`find_conflicts`

Find rules that conflict with each other within a scope or across scopes.

Parameters: scope (string, optional), rule_id (string, optional)

`discover_rules`

Scan project artifacts (configuration files, documentation) to discover implicit rules.

Parameters: file_paths (array of strings), repository (string, optional)

Returns a list of candidate rules discovered from the provided files, with confidence scores and suggested metadata.

Evaluation and Compliance

`evaluate_compliance`

Evaluate a code change or action against applicable rules.

Parameters: diff (string), file_paths (array of strings, optional), intended_action (string, optional), scope (string, optional), facts (string, optional -- JSON key-value context), environment (string, optional -- when provided, evaluates against the snapshot deployed to this environment instead of the live rule corpus; valid values: production, staging, development)

Governance

`create_proposal`

Create a governance proposal for rule changes (create, amend, retire, merge, split, override). Proposals go through draft, review, approval, and enactment stages.

Parameters: type (string), title (string), description (string), rule_id (string, optional), changes (object, optional)

`get_proposal_status`

Check the current status, votes, and comments on a governance proposal.

Parameters: proposal_id (string)

`register_agent`

Register an AI agent with the governance system. Enables personalized rule delivery, trust level progression, and governance participation. Agents register with a classification clearance level that determines which rules they can access.

Parameters: agent_id (string), agent_type (string), capabilities (array of strings, optional), clearance (string, optional -- one of public, internal, confidential, restricted; defaults to internal)

`get_personalized_rules`

Get rules personalized to the agent's history and current task. Mastered rules are suppressed, and historically-violated rules are weighted higher.

Parameters: agent_id (string), file_paths (array of strings, optional), task_context (string, optional)

`challenge_verdict`

Challenge a verdict the agent disagrees with. Creates an audit trail and may trigger rule improvements if similar challenges accumulate.

Parameters: verdict_id (string), reason (string), proposed_verdict (string, optional)

`request_exception`

Request a formal exception to a rule for a specific context. If similar exceptions are requested frequently, the system may auto-draft a rule amendment proposal.

Parameters: rule_id (string), context (string), justification (string)

Cross-Organizational

`evaluate_subject`

Evaluate any subject kind (code diff, contract clauses, HR event, transaction, etc.) against applicable rules using the polymorphic evaluation pipeline.

Parameters: subject_kind (string), payload (object), facts (object, optional), mode (string, optional)

`list_available_surfaces`

List all registered evaluation surfaces with their descriptions and supported subject kinds.

Parameters: none

`lookup_norm_lineage`

Trace a rule's derivation chain upstream to its source law/regulation, or downstream to all derived operational rules.

Parameters: rule_id (string), direction (string — upstream or downstream), max_depth (integer, optional)

`find_clause_conflicts`

Analyze a contract for clause-level conflicts against organizational standard clause rules.

Parameters: contract_text (string), contract_type (string, optional)

`check_action`

Evaluate a human action (overtime registration, leave request, expense submission) against applicable rules.

Parameters: actor (string), action (string), payload (object)

`review_communication`

Review an outbound communication (email, Slack message) for compliance with communication policies.

Parameters: channel (string), content (string), recipient_type (string, optional)

Domain-Specific Rule Retrieval

`get_rules_for_contract_review`

Get applicable rules for reviewing a contract. Primary tool for legal review agents.

Parameters: contract_type (string), parties (array, optional), governing_law (string, optional), language (string, default "ja"), max_rules (integer, default 15), format (string, default "instructions")

`get_rules_for_transaction`

Get applicable rules for validating a business transaction. Primary tool for finance/HR automation agents.

Parameters: transaction_type (string), amount (number, optional), department (string, optional), actor_role (string, optional), max_rules (integer, default 15), format (string, default "instructions")

`get_rules_for_communication`

Get applicable rules for reviewing communications. Primary tool for content review and compliance agents.

Parameters: channel (string), audience (string, optional), content_type (string, optional), max_rules (integer, default 15), format (string, default "instructions")

Domain-Specific Evaluation

`evaluate_contract`

Evaluate a contract or contract clause against applicable rules. Returns verdicts with clause-level remediations.

Parameters: contract_text (string), contract_type (string, default "other"), language (string, default "ja"), focus_areas (array, optional)

`evaluate_transaction`

Evaluate a business transaction against applicable rules. Returns verdicts with field-level remediations.

Parameters: transaction_payload (string — JSON), transaction_type (string, default "other"), actor_role (string, optional), department (string, optional)

`evaluate_communication`

Evaluate a communication draft against applicable rules. Returns verdicts with text-level remediations.

Parameters: text (string), channel (string, default "email"), audience (string, default "external"), language (string, default "ja")

Resources

URI Pattern	Description
`rule://{id}`	A single rule by its ID
`ruleset://{scope}`	All rules within a given scope

Prompts

The MCP server exposes 3 prompt templates:

compliance_check -- Systematic compliance evaluation against applicable rules
rule_summary -- Executive summary of rules by scope
impact_analysis -- Proposed change impact assessment

Configuration

Claude Code

Add the following to your Claude Code MCP configuration (.claude/mcp.json or project settings):

{
  "mcpServers": {
    "rule-repository": {
      "command": "uv",
      "args": [
        "run",
        "--directory", "/path/to/rule-repository/apps/server",
        "rulerepo-mcp"
      ],
      "env": {
        "RULEREPO_SERVER_URL": "http://localhost:8000"
      }
    }
  }
}

For remote mode (streamable-http), point your MCP client to http://localhost:8001/mcp.

Docker Compose

The mcp-server service is included in docker-compose.yml and listens on port 8001:

mcp-server:
  build:
    context: .
    dockerfile: infra/docker/Dockerfile.server
  command: ["rulerepo-mcp", "--transport", "streamable-http", "--port", "8001"]
  ports:
    - "8001:8001"
  environment:
    - RULEREPO_SERVER_URL=http://server:8000

Example: Agent Using get_rules_for_context

When an agent is about to edit files, it can call get_rules_for_context to retrieve applicable rules:

Tool: get_rules_for_context
Input: {"file_paths": ["src/api/routes.py", "src/services/auth.py"], "format": "instructions"}

Output:
- All public API functions must have type hints and Google-style docstrings.
- Authentication changes require review from the security team owner.
- Never raise bare Exception; use the project exception hierarchy.

Example: Agent Governance Flow

A coding agent can register, receive personalized rules, and challenge verdicts:

1. register_agent(agent_id="claude-code-1", agent_type="coding_assistant")
2. get_personalized_rules(agent_id="claude-code-1", file_paths=["src/api/auth.py"])
   -> Returns rules tailored to this agent's history (mastered rules suppressed)
3. evaluate_compliance(diff="...", file_path="src/api/auth.py")
   -> DENY on rule-42
4. challenge_verdict(verdict_id="v-123", reason="Rule does not apply to internal endpoints")
   -> Challenge recorded; if pattern accumulates, may auto-draft rule amendment

Tool count accuracy note (2026-05-15): The 24 tools listed above have been verified against the async def function definitions in apps/server/src/rulerepo_server/mcp/tools.py. If tools are added or removed in the code, update this document accordingly.

MCP Server

Transport Modes

Tools

Search and Context

search_rules

explain_rule

get_rules_for_context

find_conflicts

discover_rules

Evaluation and Compliance

evaluate_compliance

Governance

create_proposal

get_proposal_status

register_agent

get_personalized_rules

challenge_verdict

request_exception