ofrag

ofrag is a specialized Retrieval-Augmented Generation (RAG) engine designed to enhance LLMs with local schema awareness and live data access to Oracle Fusion Cloud. It combines local semantic tooling (lint_sql, semantic_search, search_identifiers, module_summary) with an execution layer that runs validated SQL against your live Oracle Fusion instance. The result is an AI assistant that can understand user intent, validate and repair SQL against the Oracle schema locally, then run the query live and present results in a clean, formatted markdown table. This enables rapid, self-contained analysis and data retrieval without repeatedly jumping through BI tools.

To use it, run the OFRAG MCP server and interact with it through your preferred LLM integration. When you ask questions like “show me top invoices,” the system can lint and validate the SQL, ensure the target tables exist (e.g., AP_INVOICES_ALL), and execute the query against Oracle Fusion. The local tools provide precise feedback about schema structure and potential issues, which the LLM can then incorporate into its reasoning and repairs before executing the final query.

Prerequisites

• A supported environment with Node.js installed (or a containerized/runtime that can execute Node-based MCP servers).
• Access to an Oracle Fusion Cloud instance with credentials configured for live queries.
• Local schema cache and metadata available to OFRAG (as described in the project docs).

Install steps

1. Install Node.js if you don’t have it already. Visit https://nodejs.org/ and follow the installation instructions for your OS.

2. Clone the MCP repository or download the OFRAG release package.

3. Install dependencies (if applicable to the project layout). Example:

   npm install

4. Configure environment variables and defaults (see environment variables in the mcp_config section). Create a config file if required by your setup, for example config/ofrag-config.json.

5. Start the MCP server using your preferred method (as defined by the mcp_config). Example if you’re using a direct Node.js invocation:

   node path/to/server.js

6. Verify the server is running and accessible via the MCP gateway/interface and that it can reach the Oracle Fusion Cloud environment with the configured credentials.

Tips and notes:

• Ensure the local Oracle Fusion metadata cache is up to date so lint_sql and semantic tools validate queries correctly.
• Use the lint_sql tool to catch structural issues before executing queries against live data.
• Review the generated SQL feedback from semantic_search and module_summary to understand any suggested changes to queries.
• If you encounter connection issues to Oracle Fusion Cloud, verify network rules, VPNs, and the credentials configured in OFRAG.
• Set the OFRAg_LOG_LEVEL to debug during troubleshooting to capture more verbose logs.
• If you plan to run multiple servers, isolate their environments and ensure unique namespace/config paths to avoid clashes.