Skip to main content
Version: v1.0.0

Pharus MCP Server

Pharus ships a hosted MCP (Model Context Protocol) server, so AI agents — Claude, Claude Code, or any MCP-capable agent — can operate your Pharus organization directly: no HTTP plumbing, no remembered endpoints, just typed tools.

Your agent  ──MCP──▶  https://stg.api.pharus.ai/mcp  ──▶  Pharus API
                       (OAuth or a PAT as a Bearer token)

What agents can do​

The server exposes the full data-plane API as tools (144 of them — see the Tool Reference):

  • Master data — materials, SKUs, BOMs, vendors, customers, brokers, locations, contacts
  • Flows — purchase orders, work orders, sales orders: create, update, line items, receipts
  • Inventory — stock summaries, ledgers, adjustments, transfers, disposals for both materials and finished goods
  • Documents — list, read (signed download URLs), delete

Typical asks an agent can handle end-to-end: "How many units of SKU X do we have in the Bilbao warehouse?", "Create a PO to restock everything below its reorder point", "Receive yesterday's delivery against PO-0042", "Move 200 units to the new location and complete the transfer when it arrives".

What stays in the web client​

The control plane: organization settings, members, invitations, and token management are not exposed as tools. Humans manage the platform; agents operate the data.

Safety model​

  • Your credentials, your permissions. Every tool call runs through the exact same API auth as any other client: the connection's organization binding and your role in that organization (member/admin/owner) are enforced server-side per call — whether you connect over OAuth or with a PAT.
  • Destructive tools are flagged. Deletes and disposals carry MCP destructive annotations, so well-behaved agent hosts ask for human confirmation first.
  • No drift, ever. The tool surface is generated from the same schema definitions as the HTTP API and API reference; CI fails if an endpoint is added without a corresponding tool (or an explicit exclusion).

Get started​

  1. Authenticate — connect over OAuth (Claude Desktop, claude.ai) or create a personal access token (Claude Code, the API, other clients)
  2. Connect your agent
  3. Browse the Tool Reference