Agent API

Give AI assistants and your own tools a read-only window into production planning — the data needed to answer “what should we make today, and will any shipments slip?” Connect over the Model Context Protocol for Claude, wire it into a custom GPT, or call a bearer-token API directly. Ashicore exposes context; it never lets an assistant change your workspace.

Access: read-onlyManage: Owner or settings adminSettings → Agent API

What it exposes

The agent surface serves read-only planning data over two tools — a production-planning context and a replenishment context. Unit costs are stripped out before anything leaves your workspace. Together they provide a snapshot of everything a scheduler needs, and nothing it doesn’t:

Open sales ordersOpen manufacturing ordersProduct counts on handLot countsBOM requirements

Read-only by design

This is the external planning surface — it reads production context and returns answers. It does not create orders, move stock, or change anything in Ashicore. On-screen, the app states it plainly: “Credentials are read-only in v1 — assistants can look things up but can’t change your data.” The in-app agent chat panel is a separate, currently-disabled surface.

Ways to connect

A

Claude — over MCP

An OAuth endpoint (PKCE, with a fixed set of redirect URIs — not a pasted token). From Settings → Agent API, choose Connect Claude — it opens Claude with the Ashicore connector pre-filled. Authorizing requires planning-read access.

B

ChatGPT — a custom GPT action

Use Setup guide and Copy schema URL to add a custom GPT action backed by a bearer token and the published OpenAPI schema.

C

Your own tools — a bearer-token API

Mint a bearer API token (shown once, then listed by name) with the planning-read scope and call the planning context endpoints directly. An OpenAPI description is published, so you can wire it into scripts, dashboards, or a custom GPT. Managing tokens is a team-manager task.

The remote MCP server lives at a single URL:

https://ashicore.app/api/agent/mcp

Setting it up

1. Open Settings → Agent API. You’ll need to be the workspace owner or a team manager.

2. For Claude (Method A), choose Connect Claude — this starts the OAuth/PKCE flow. For ChatGPT (Method B), use Setup guide and Copy schema URL.

3. For direct access, choose Create API token and store it as a bearer secret in your tool — the token is shown once.

4. Review Active credentials any time. Granted MCP authorizations and API tokens are listed together, and you can Revoke either.

The Ashicore Claude plugin

Alongside the raw connection, Ashicore ships a small Claude plugin that adds a Production Schedule skill on top of the MCP server. Ask Claude what to make each day, or whether any shipments are at risk, and it pulls your live planning context first, then returns a clear daily schedule — explicitly calling out any shipment that will be missed and what would have to happen, on what day, to save it.

It won't invent capacity

If the schedule needs a daily capacity you haven’t provided, the skill says so rather than guessing — the planning stays grounded in your real production data.

Related

• Prioritize the queue — the open-order ranking the schedule reads from.
• Fulfillment & allocation — how demand and coverage are computed across orders.
• Integrations overview — all connections at a glance.