Ordio MCP

Ordio MCP

MCP server for the Ordio restaurant operations platform, enabling AI assistants to manage inventory, orders, recipes, team, shifts, and reports.

Category
Visit Server

README

Ordio MCP

Ordio MCP — Model Context Protocol server exposing Ordio data and actions to AI assistants.

Ordio MCP turns the Ordio restaurant operations platform into a set of structured tools, resources, and prompts that any AI assistant can use. Ask Claude (or any MCP-capable agent) to check inventory, draft a purchase order, schedule a shift, or run a variance report — and it can do so safely, scoped to your organization.

What is MCP?

The Model Context Protocol is an open standard for connecting AI assistants to external systems. An MCP server exposes capabilities (tools, resources, prompts) over a simple JSON-RPC interface; an MCP client (such as Claude Desktop, Claude Code, or Windsurf) connects to the server and lets the model call those capabilities on the user's behalf.

Highlights

  • ~145 tools spanning inventory, recipes, vendors, orders, invoices, tasks, team, shifts, alerts, kitchen board, AI extraction, reports, and analytics
  • 5 organization resources for at-a-glance context (profile, inventory summary, active alerts, team summary, daily metrics)
  • 6 prompt workflows (inventory audit, purchase-order workflow, shift scheduling, end-of-day closing, recipe costing, new menu-item setup)
  • stdio and SSE transports
  • Strict per-request org scoping and Bearer-token auth against the Ordio REST API
  • Two tool sources, selectable at startup: the hand-written tool modules (default) or the unified assistant catalog proxy (ORDIO_CATALOG_PROXY=1), which routes every call through the Ordio API's chat/voice gateway so the MCP and in-app assistant surfaces can never drift — see Tool source

Available tools

Tools are grouped by domain. The full list lives in src/tools/.

Group Count Examples
Inventory 6 list_inventory, create_inventory_item, batch_update_inventory
Recipes 5 list_recipes, create_recipe, update_recipe
Recipe batches 5 list_recipe_batches, create_recipe_batch
Vendors 5 list_vendors, create_vendor
Categories 5 list_categories, create_category
Purchase orders 5 list_orders, create_order
Invoices 5 list_invoices, create_invoice
Tasks 5 list_tasks, create_task
Team 9 list_team_members, list_departments, create_department
Shifts 5 list_shifts, create_shift
Shift requests 5 list_shift_requests, create_shift_request
Alerts 9 list_alerts, dismiss_alert, create_alert_configuration
Menu items 5 list_menu_items, create_menu_item
Kitchen board 14 list_kitchen_orders, bump_order, list_routing_rules
Count sessions 5 list_count_sessions, create_count_session
Storage locations 5 list_storage_locations, create_storage_location
Units 5 list_units, create_unit
Sales & labor 8 list_sales_data, list_labor_daily_summary, list_timesheets
Activity 3 list_activity, log_activity
Settings 4 get_settings, upsert_setting
Reports 3 get_variance_report, get_ap_aging_report, get_cost_analysis
Temperature logs 2 list_temperature_logs, create_temperature_log
Equipment 4 list_equipment, create_equipment
AI & automation 11 ai_process_invoice, ai_extract_recipes, ai_generate_recipe
Search 1 semantic_search
Analytics 6 inventory_health_dashboard, food_cost_analysis, daily_operations_briefing

Resources

URI Description
ordio://org/profile Organization profile and settings
ordio://org/inventory-summary Total items, low stock, expiring
ordio://org/active-alerts Currently active/unresolved alerts
ordio://org/team-summary Team count and today's schedule
ordio://org/daily-metrics Today's alerts, tasks, inventory health

Prompts

inventory-audit, purchase-order-workflow, shift-scheduling, end-of-day-closing, recipe-costing, new-menu-item-setup.

Install and configure

Claude Code

claude mcp add ordio \
  --env ORDIO_API_KEY=sk_live_... \
  --env ORDIO_ORG_ID=org_xxxxxxxxxx \
  -- node /absolute/path/to/ordio-mcp/dist/index.js

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "ordio": {
      "command": "node",
      "args": ["/absolute/path/to/ordio-mcp/dist/index.js"],
      "env": {
        "ORDIO_API_KEY": "sk_live_...",
        "ORDIO_ORG_ID": "org_xxxxxxxxxx"
      }
    }
  }
}

Restart the client to pick up the new server.

Build first

git clone https://github.com/Abhilashachar/ordio-mcp.git
cd ordio-mcp
npm install
npm run build

Authentication

Every request to the Ordio REST API is sent with Authorization: Bearer ${ORDIO_API_KEY} and is scoped to ORDIO_ORG_ID. You'll need:

Variable Required Default Description
ORDIO_API_KEY yes — Clerk-issued API key used as the Bearer token
ORDIO_ORG_ID yes — Organization ID; all tool calls are scoped to this org
ORDIO_API_BASE_URL no https://api.getordio.com Override for staging or self-hosted deployments
ORDIO_MCP_TRANSPORT no stdio stdio or sse
ORDIO_MCP_PORT no 3100 Port for the SSE HTTP server
ORDIO_CATALOG_PROXY no off Set to 1/true to register tools from the API assistant catalog (proxy) instead of the hand-written tool files

Treat your API key like a password — it grants full access to the configured organization.

Tool source: hand-written vs. catalog proxy (ORDIO_CATALOG_PROXY)

The server can expose its tools from one of two sources, selected at startup by the ORDIO_CATALOG_PROXY flag:

  • Default (flag off): the 27 hand-written tool modules under src/tools/ are registered (the ~145 tools listed above). This remains the default during the transition — leaving the flag unset changes nothing.
  • Catalog proxy (ORDIO_CATALOG_PROXY=1): the server fetches the ordio-api assistant tool catalog at startup (GET /api/v1/orgs/:orgId/assistant/tools) and registers one MCP tool per catalog entry, executing each through the API's chat/voice gateway (permission scoping, confirmation gating, write rate-limits, and idempotency all apply). This makes the MCP surface and the in-app chat/voice surface the same catalog, so they cannot drift, and a new capability added to the catalog appears here automatically. If the manifest fetch fails — or a single entry is malformed — the proxy degrades gracefully (that tool, or all tools, are skipped) rather than crashing the server.

Why it isn't the default yet. The two sources are mutually exclusive, and the hand-written files are retained for one more step. The assistant catalog is now a superset of this server's non-destructive tools — the domains that used to be MCP-only (oracle, reports, equipment, temperature logs, station/"department" writes, menu-item writes, alert lifecycle, time-clock, kitchen writes, …) are all in the catalog, so flipping the flag no longer drops them. The one deliberate difference is destructive operations: the assistant catalog excludes hard delete_* tools by policy (the assistant never destroys data — use reversible archive/restore instead), so those live only in the hand-written set. Once the proxy is validated against the live manifest in staging, a follow-up flips the default on and removes the hand-written modules.

Prerequisites: the API's /assistant/tools endpoint must be deployed (it is, on ordio-api main), and ORDIO_API_KEY must resolve to a user whose permissions cover the tools you want exposed — the manifest is permission-filtered per caller, and every write still passes the gateway's confirmation/rate-limit checks. (A durable service-user token minted via Clerk, rather than a hand-pasted key, is a planned follow-up.)

Transports

  • stdio (default): the standard MCP transport used by Claude Desktop, Claude Code, and Windsurf.
  • SSE: an HTTP server with Server-Sent Events for remote or browser-based clients. Enable with ORDIO_MCP_TRANSPORT=sse.

Local development

npm install
cp env.example .env       # fill in ORDIO_API_KEY and ORDIO_ORG_ID
npm run dev               # tsx watch — no build step
npm run typecheck         # tsc --noEmit
npm run lint              # eslint src/
npm test                  # vitest

Source layout:

src/
  index.ts          MCP server entrypoint + tool registration
  config.ts         Environment loading
  client.ts         Thin REST client (Bearer auth, org scoping)
  tools/            One module per domain (inventory, recipes, ...)
  resources/        Organization resources
  prompts/          Workflow prompts
  transports/sse.ts SSE HTTP transport
  utils/            Error and response helpers

Status

Active. Version 0.2.0. APIs and tool names may still evolve before a 1.0 release. The server is mid-migration from hand-written tool modules to the unified catalog proxy; the default remains the hand-written set until the proxy is validated in staging. Issues and contributions are welcome.

License

Released under the MIT License.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured