AI & Agents · MCP servers for Oracle ERP

We connect Claude to Oracle EBS and Fusion Cloud. And we use it ourselves, every day.

Our MCP servers let an AI agent deploy a verified package, check whether a git branch matches what's installed in TEST, query the database, or manage the BI Publisher catalog. This isn't a slide deck: it's the tooling our own consultants run on real EBS R12.2 environments.

Book a call ▶ Watch a real session

mcp — ebs-r12 · 4 servers connected● healthy

katrina.deploy_package — gated, auditedkatrina.verify_branch — git vs environmentkatrina.rollback — requires approvalebs.query — read-only SQLebs.concurrent_status — requests, logsbip.catalog — list, diff, upload templatesfusion.fbdi_load — requires approvalPROD: blocked without change ticket — all servers

What an agent can do

Real tools against real environments — not a chatbot on your docs.

katrina.deploy_package

Deploy a verified package

The agent requests a deploy through Katrina — same installer, same approval gates, same audit trail as a human. It can't skip a gate because there is no path around it.

katrina.verify_branch

Check branch vs environment

"Does TEST actually have what's in feature/xxar-aging?" — answered in seconds, from installed build metadata, not from someone's memory.

ebs.query · ebs.concurrent_status

Query the database and concurrent managers

Read-only SQL against EBS, request statuses, log retrieval. The agent diagnoses a stuck concurrent request the way a support analyst would — but at 3 AM, in seconds.

bip.catalog · fusion.fbdi_load

Manage BI Publisher and Fusion loads

List, diff and upload BI Publisher templates; stage FBDI loads on Fusion Cloud. Hybrid landscapes get one agent that speaks to both systems.

Recorded session

Claude Code deploying to an EBS environment. For real.

Ninety seconds, unedited: the agent verifies the branch against TEST, requests the deploy, hits the approval gate, and reports the audit log entry. Nobody else can show you this.

Recorded on a real R12.2.12 instance. Client identifiers redacted.

$ claude "is feature/xxar-aging installed in TEST? if so, promote to UAT"⏺ katrina.verify_branch(branch: "feature/xxar-aging", env: "TEST") ⎿ installed build 3812.4 = branch HEAD ✓ match⏺ katrina.deploy_package(id: 3812, env: "UAT") ⎿ approval gate: waiting for release manager… ⎿ approved — f.martinez · 14:32 ART ⎿ installer running: 14/14 objects OK✓ package 3812 installed on UAT — audit log #48237

▶0:31 / 1:28

Guardrails

An ERP is not a playground. The agent gets exactly the permissions a junior consultant would.

Every action logged

Full trace of every tool call — inputs, outputs, who triggered it — in the same audit trail as human activity.

Destructive ops require approval

Deploys, rollbacks and loads stop at a human gate. Read-only tools run free; anything that writes, waits.

PROD blocked without a ticket

Hard-coded at the server level, not the prompt level. No change ticket, no production — for agents and humans alike.

The rails underneath

Katrina — CI/CD for EBS

The agent's deploys ride Katrina's gates and audit trail. See Katrina →

Hybrid landscapes

EBS + Fusion, one agent

Agents that work across both systems — because we run both. Oracle EBS → · Fusion & OCI →

Start with one workflow, on your environment.

A pilot agent on a non-production instance — deploy verification, support triage, or BI Publisher management. You see the audit log from day one.

Book a 30-minute call

With the engineers who wrote the MCP servers.

Book a call

We reply within one business day, ET/ART.