Give your AI agent direct access to Oracle Fusion ERP data. Ask questions in plain English — get SQL results in seconds.
Demo · What is it · Tools · Quick Start · Docs
Oracle Fusion Cloud has 25,000+ tables across Finance, SCM, HCM, and Procurement. Finding the right table, understanding join relationships, and writing correct SQL takes hours of digging through Oracle docs. Even experienced consultants spend 15+ minutes per query cycle: log into BI Publisher → create data model → write SQL → debug → re-run.
ofrag eliminates this entirely. Your AI agent discovers tables semantically, validates SQL against a local schema cache, and executes queries against live Fusion data — all in one conversation.
You: "Show me the top 5 unpaid invoices for vendor Acme Corp generated last month."
Claude + ofrag:
- Discovers the right tables via semantic search (
AP_INVOICES_ALL,POZ_SUPPLIERS,HZ_PARTIES)- Understands "unpaid" means
PAYMENT_STATUS_FLAG = 'N'- Validates the SQL against local metadata — catches errors before they hit production
- Executes the query against your live Oracle Fusion instance
- Returns a formatted table with invoice numbers, amounts, and dates
Total time: ~4 seconds.
No BI Publisher. No data model. No OTBI subject areas. Just a question and an answer.
ofrag is an MCP (Model Context Protocol) server — an open standard that lets AI agents use external tools. It gives Claude, OpenAI, Gemini, and any MCP-compatible LLM structured access to your Oracle Fusion Cloud data.
┌──────────────────────────────────────────────────────────────┐
│ Your AI Agent │
│ Claude · OpenAI · Gemini · Claude Code │
├──────────────────────────────────────────────────────────────┤
│ ofrag (MCP Server) │
│ │
│ ┌─────────────┐ ┌──────────────┐ ┌─────────────────────┐ │
│ │ Semantic │ │ SQL Linter │ │ Business Process │ │
│ │ Search │ │ (AST-based) │ │ Mapper │ │
│ └──────┬──────┘ └──────┬───────┘ └──────────┬──────────┘ │
│ │ │ │ │
│ ┌──────▼──────────────────────────────────────▼──────────┐ │
│ │ Local DuckDB Metadata Cache │ │
│ │ 25,000+ tables · 240K+ descriptions │ │
│ └────────────────────────┬───────────────────────────────┘ │
│ │ Only validated queries │
├───────────────────────────▼──────────────────────────────────┤
│ Oracle Fusion Cloud │
│ BI Publisher SOAP · REST APIs │
└──────────────────────────────────────────────────────────────┘
Key principle: All schema exploration happens locally — zero queries to production during discovery. Only final, validated SQL reaches your live Fusion instance.
| ofrag + AI Agent | Manual BI Publisher | OTBI | Proprietary SQL Tools | |
|---|---|---|---|---|
| Natural language queries | ✅ Ask in plain English | ❌ Write SQL manually | ❌ Drag-and-drop only | ❌ Write SQL manually |
| Schema discovery | ✅ Semantic search across 25K+ tables | ❌ Read Oracle docs | ❌ Limited subject areas | ✅ Table browser |
| SQL validation before execution | ✅ AST parsing against local cache | ❌ Trial and error | N/A | ❌ |
| Cross-module table relationships | ✅ AI traces AP→XLA→GL chains | ❌ Manual knowledge | ❌ Single subject area | ❌ |
| Business process mapping | ✅ business_process_map, scenario_mapper |
❌ | ❌ | ❌ |
| Multi-environment switching | ✅ DEV/SIT/UAT/PROD in one session | ❌ Separate logins | ❌ Separate logins | Partial |
| REST API access | ✅ Built-in universal REST client | ❌ | ❌ | ❌ |
| Production safety | ✅ Local-first, validated queries only | ✅ Sandboxed |
The real difference: Other tools help you write SQL. ofrag lets you skip writing SQL — the AI does it for you, correctly, using verified metadata.
ofrag exposes 30+ tools to your AI agent, organized by function:
| Tool | What it does |
|---|---|
semantic_search |
Find tables/columns by business meaning ("unpaid invoices", "employee compensation") using vector embeddings across 240K+ descriptions |
search_identifiers |
Find tables/columns by Oracle name pattern (AP_INVOICES, VENDOR_ID) |
search_descriptions |
Exact keyword search in table/column descriptions |
list_tables |
Browse tables by module with filters |
list_columns |
Get all columns, types, and descriptions for a table |
index_info |
Index definitions for a table |
table_overview |
Quick summary: column count, purpose, module |
| Tool | What it does |
|---|---|
execute_oracle_sql |
Execute live SQL against Oracle Fusion, results inline |
execute_oracle_sql_to_file |
Execute SQL, save results to local file (for large datasets) |
lint_sql |
Parse SQL into AST, validate against cached schema — catches errors before execution |
suggest_sql |
Auto-complete partial SQL using metadata cache |
raw_select |
Query the local DuckDB metadata cache directly |
| Tool | What it does |
|---|---|
module_summary |
Overview of a Fusion module — purpose, key tables, typical queries |
business_process_map |
Map a business process to its tables, stages, and key columns |
scenario_mapper |
Map a business intent ("month-end close") to recommended tables and metrics |
relationship_map |
Show inbound/outbound JOIN relationships for any table |
cross_module_analyzer |
Analyze how a table connects across Fusion modules |
integration_flow_mapper |
Trace end-to-end data flow: upstream feeders → table → downstream consumers |
process_catalog |
Full business process catalog with all stages and tables |
| Tool | What it does |
|---|---|
rest_call |
Make HTTP REST API calls to Oracle Fusion, results inline |
rest_call_to_file |
REST API calls with results saved to file |
| Tool | What it does |
|---|---|
describe_flexfield |
Describe Descriptive Flexfields (DFF) — contexts, segments, value sets |
lookup_values |
Get all values for a Fusion lookup type (LOV) |
profile_values |
Get profile option values at site/product/user levels |
| Tool | What it does |
|---|---|
run_bi_report |
Execute a BI Publisher report with parameters, results inline |
run_bi_report_to_file |
Execute BI Publisher report, save output to file |
submit_ess_job |
Submit an ESS job (batch processes, imports, exports) |
get_ess_job_status |
Check ESS job status |
| Tool | What it does |
|---|---|
authenticate |
Browser-based SSO login to Oracle Fusion |
get_auth_status |
Check current authentication state |
add_environment |
Add/update a named environment (DEV, SIT, UAT, PROD) |
switch_environment |
Switch active environment instantly |
list_environments |
List all configured environments |
compare_environments |
Compare configuration data between two environments |
Download from Releases (Windows and macOS).
👉 Installation & Setup Guide — step-by-step: licensing, BI Publisher report deployment, MCP client configuration (Claude Desktop, Gemini, Claude Code), authentication setup.
"What tables store AP invoice data?"
"Show me all overdue invoices for Acme Corp"
"How are accounts payable balances linked to the general ledger?"
"Compare the supplier count between DEV and PROD"
Ad-hoc data analysis — Ask "show me revenue by business unit for Q4" and get results in seconds. No OTBI report creation, no BI Publisher data model, no waiting.
Implementation troubleshooting — During Fusion implementation, quickly verify data: "are there any AP invoices without distributions?", "show me GL journal entries that haven't been transferred from XLA".
Data reconciliation — "Compare AP subledger totals against GL balances for period DEC-24". The AI traces the full AP → XLA → GL chain and identifies discrepancies.
Schema exploration — "What tables store employee compensation data?" The AI uses semantic search to find relevant HCM tables, shows their relationships, and explains the data model.
Integration development — Building OIC integrations? Ask "what REST endpoints exist for AP invoices?" or "what's the FBDI template for importing journals?" — ofrag finds it.
EBS → Fusion migration — "What's the Fusion equivalent of AP_INVOICES_ALL in EBS?" The AI maps EBS tables to Fusion tables using semantic search and module knowledge.
Report development — Before building a BI Publisher report, prototype your SQL through conversation. The AI validates and optimizes your queries using the local schema cache.
- Read-only SQL —
SELECTqueries only through BI Publisher SOAP. Oracle Fusion does not permit write access through this layer. - Security — Ensure usage complies with your organization's security policies. Credentials are stored locally and never transmitted to third parties.
| Guide | Description |
|---|---|
| Installation & Setup | Detailed guide on how to install, configure, and get the binary running |
| Full MCP Server Reference | Complete reference for configuration, authentication, variables and tools |
| Multi-Environment Management | Configure and switch between dev, UAT, prod environments interactively |
| Semantic Search Setup | How to enable vector search and REST API catalog for full-powered semantic discovery |
ofrag is part of the Oracle Fusion open-source ecosystem:
| Project | What it does | Link |
|---|---|---|
| ofrag | MCP Server — AI-powered queries via Claude, OpenAI, Gemini | GitHub |
| OFJDBC | JDBC driver — SQL access from DBeaver, IntelliJ, JVM apps, ETL pipelines | GitHub |
- Website: oraclefusionsql.com
- GitHub Issues: krokozyab/ofrag/issues
- Email: sergey.rudenko.ba@gmail.com
- LinkedIn: Sergey Rudenko
If ofrag saved you time, consider leaving a ⭐
The only MCP server for Oracle Fusion Cloud. Built by an Oracle Fusion consultant who got tired of BI Publisher.