Fusion BIP MCP Server

An MCP server that gives an AI agent safe, described access to Oracle Fusion BI Publisher. Instead of handing an agent raw credentials, you expose a small set of scoped tools it can call — run a query, run a report, fetch or deploy catalog objects — and the server talks to Fusion's BI Publisher SOAP service directly. No database and no middle tier to stand up.

This is a public, source-only mirror of the core server for people who want to read the code or point their own agent at their own Fusion instance.

The tools

The destructive delete_catalog_object tool stays unregistered unless you explicitly opt in with FUSION_MCP_ENABLE_DELETE=1, so an agent cannot delete anything it was never given.

Related MCP server: ONE-MCP

How it fits together

AI agent (Claude Code, etc.)
      │  MCP (stdio)
      ▼
  server.py        registers the tools, meters usage
      │
  client.py ─────► direct_client.py ──► BI Publisher SOAP  ──►  Oracle Fusion
      │                                   (soap.py builds the envelopes,
  context.py                               execute.py runs them,
  (credentials/targets)                    builders.py shapes .xdm/.xdo)

• server.py — the MCP server; registers each tool and its description.

• client.py — the backend seam the server imports.

• direct_client.py — the six operations implemented against BI Publisher SOAP.

• soap.py / execute.py / builders.py — envelope construction, execution, and .xdm/.xdo artifact building.

• context.py — where the Fusion pod and user credentials are resolved for a call. The defaults in this mirror are placeholders (your-pod, bip_user); wire in your own.

• metering.py — an append-only, metadata-only usage log.

Running it locally

pip install -e ".[stdio]"    # installs the mcp runtime
# set your Fusion pod + user in context.py (or via environment) and register
# fusion-mcp with your MCP client (Claude Code / Desktop) over stdio.

Notes

• This mirror is the direct-over-SOAP path only. Credential wiring is intentionally left as placeholders — it does not run as-is against any real instance until you supply your own Fusion pod and a user with the BI Publisher Author role.

• No credentials, tokens, or private endpoints are included.

License

MIT — see LICENSE.