MCP connector

Ask Claude about your docks.

Dock Optimizer speaks the Model Context Protocol — the open standard AI agents use to reach real systems. Connect Claude and it can read your organization's dock schedule, inventory, loads and partner relationships, then answer in plain language. Read-only by default. One consent screen to set up.

Set it up How access is controlled

Works with claude.ai, Claude Desktop, and any MCP client
https://dockoptimizer.com/api/mcp/v1/message

What it can read

Your whole operation, one question away

A connected agent sees the same data your team sees in the app — 59 read resources, scoped to your organization, each one paginated and bounded so answers stay fast.

Docks and appointments

Appointments by day, with check-in and check-out state
Door status, availability and booking history
Slot availability per appointment type
No-show, stale and pending appointments

Inventory, to the transaction

Stock levels, low-stock and quarantine flags
Every receive, pull and adjustment, with lot numbers
Reservations against sales orders
Storage locations, kitting recipes, cycle counts

Freight and partners

Loads, bids, sales and purchase orders
Carrier performance and lane analytics
What you hold for customers — and what partners store for you
Detention, accessorial charges, statements

In practice

Questions operators actually ask

When is my dock busiest, by weekday and hour?Answered from the same data behind the heat map.

Which inbound appointments tomorrow are missing a PO number?Appointments carry their check-in answers, so gaps are visible.

How much stock do we hold for this customer, and how much is reserved?Inventory is owner-aware — a 3PL's customer sees exactly their own stock.

Which carriers no-show most at our facilities?Arrival performance from dock check-in and check-out records.

You: What's on the dock schedule Monday?
Claude: 38 appointments. The 8–10am window carries 14 of them —
9 at 450 Airtech, mostly LTL drop-offs. Two arrivals have no PO
on file: #17452 (returns) and #17466. Want the full list?

Security model

Built for the org boundary

Warehouses hold other companies' inventory and schedules. The connector is designed around that fact: every grant is scoped to one organization, resolved server-side, and nothing about a request can widen it.

Org-scoped by constructionConsent comes from your org's admin, and the token it mints reads that organization only. The audience is resolved on our servers at consent time — no request parameter can name a different org.

Read-only by defaultA connected agent can read, and that's all. It advertises no write tools, and write calls are refused at the scope check — not hidden in a policy document.

Writes wait for a humanOrganizations that issue write-capable agent keys route changes through proposal tools: the agent proposes, a person approves it in Integration Hub before anything happens.

Every call auditedEach request is logged with its key, resource and outcome, and the log is visible to your admins in Integration Hub → Agent Keys.

Standard OAuth 2.1Authorization-code flow with PKCE, dynamic client registration, one-hour access tokens, rotating refresh tokens, and tokens pinned to the environment that issued them.

Rotation without breakageRotating an agent key keeps the old key working for 24 hours while you swap in the new one, so a rotation never takes down a running integration. Revocation is immediate when you want it to be.

Setup

One consent screen

Turn on Integration HubAgent access is part of the Integration Hub subscription. If you already see Integration Hub in your sidebar, you're set.

Add the connector in ClaudeIn claude.ai: Settings → Connectors → Add custom connector, then paste https://dockoptimizer.com/api/mcp/v1/message.

Sign in and authorizeClaude sends you to Dock Optimizer to sign in. An Admin or Manager at your organization reviews exactly what the token can read and approves it.

AskClaude can now answer from your live operation. Access shows up under Integration Hub → Agent Keys, where you can watch it, rotate it, or revoke it.

Consent is limited to org Admins and Managers by design — connecting an agent grants read access to operational data, so the decision belongs to someone accountable for it.

For developers

Spec-compliant, discoverable, documented

MCP endpoint	`POST https://dockoptimizer.com/api/mcp/v1/message` — JSON-RPC 2.0; resources/list, resources/read, tools/list, tools/call
Manifest	`GET https://dockoptimizer.com/api/mcp/manifest.json` — every tool and resource, generated from the live registry so it cannot drift
OAuth discovery	`/.well-known/oauth-authorization-server` and `/.well-known/oauth-protected-resource` (RFC 8414 / RFC 9728)
Registration	`POST /api/mcp/oauth/register` — RFC 7591 dynamic client registration; registration alone grants nothing
Authorization	OAuth 2.1 authorization code + PKCE S256, interactive consent, single-use 10-minute codes, rotating refresh tokens
Tool metadata	Every tool carries MCP annotations (read-only, destructive, idempotent hints) so clients can render honest trust cues
Docs	API reference · status