ai·rete·rag MCP Server is an MCP (Model Context Protocol) server for connecting ai·rete·rag's deterministic rule-based decision-making capabilities to Claude Code, Claude Desktop, or any MCP-compatible client.
Purpose The server is designed to separate "decision" from "explanation": the verdict is always produced by the Rete rule engine, with auditable and reproducible characteristics; the LLM is only responsible for explaining the reasons for the verdict, and the explanation must be based on imported policy documents, thereby remaining traceable. The rules themselves can be drafted from policy text and saved in YAML format, making them easy to maintain and version.
Main capabilities
- Provides the prompt decide_case: this is a case decision workflow for any domain. It calls get_parameters to learn the fact fields on which the rules are based, then reads the case materials—text, photos, document scans, voice notes, or video (the agent reads the media itself; the API only receives facts)—and maps these materials to the exact fields.
- When decide is called, fact_sources records which file each fact comes from, and this source information is retained in the decision's audit trail.
- When missing_parameters in the result is non-empty (that is, facts that could still change the verdict are missing), the workflow asks the user for the missing facts in order from most to least decisive, and then makes the decision again.
- The verdict always comes from the rule engine and is not affected by the LLM.
How to use The server provides two access paths:
- Connect via URL (no installation required): the server also has a hosted version, which is the only feasible way for clients such as claude.ai and Claude Desktop custom connectors that can only fill in a URL and have no static header field. Add https://ai-rete-rag.com/mcp/auth as a custom connector and approve the prompt. The connection will appear under "Settings → Connected apps", and disconnecting takes effect immediately.
- Run locally: with uv, no installation is required; uvx ai-rete-rag-mcp will pull and run the server on demand. It can also be installed as a package. For first use, log in at ai-rete-rag.com, open "Settings → API Keys" to create a key (ik_..., shown only once). If installed via pip, use -- ai-rete-rag-mcp instead of -- uvx ai-rete-rag-mcp in the configuration. In addition, you can skip installation and point directly to the hosted endpoint with the key. Claude Desktop or other clients use JSON configuration; if installed via pip, set "command": "ai-rete-rag-mcp" and remove "args".
Environment variables The server is configured using environment variables (the specific variable names are not provided for now). Without a key, you can still explore shared demo domains (such as loan, fraud, clinical, etc.), but you must comply with the free quota limits.
Applicable scenarios Suitable for rule-driven scenarios that require auditable, reproducible decisions, such as loan approval, fraud detection, clinical judgment, etc., while also wanting the LLM to provide natural-language explanations based on policy documents. Example: "Use ai·rete·rag to decide whether this loan application should be approved: credit score 645, annual income $52k, requested amount $30k." Claude will call decide(domain="loan", facts={...}) and return the verdict derived from the rules along with a plain-English explanation citing the underwriting policy.
Limitations The code content is not provided, so specific implementation details cannot be shown. Facts such as environment variable names, the specific values of the free quota, and the specific rule sets for each demo domain are not provided for now.
