The Replinish MCP server

The endpoint

The MCP server speaks the Streamable HTTP transport from the current Model Context Protocol specification (revision 2026-07-28). It accepts JSON-RPC 2.0 POSTs at the single endpoint below. The server is stateless: every request carries its own protocol version in the MCP-Protocol-Version header and in the body's _meta field, so any server instance behind the load balancer can answer any request.

https://www.replinish.app/mcp

Required request headers on every POST:

• Content-Type: application/json
• Accept: application/json, text/event-stream
• MCP-Protocol-Version: 2026-07-28
• Mcp-Method and Mcp-Name, which mirror the JSON-RPC method and the tool or resource name.

Tools

Five tools, all pure math. They import directly from web/app/lib/engine/ and never touch a database. Pass the numbers in, get the answer out.

Every argument is checked against the tool's own schema before any math runs. An argument that is missing, the wrong type, not a finite number, or outside the allowed range comes back as a tool error naming the parameter, so the model can correct itself and try again.

reorder_point

The stock level at which you have to place a new order to avoid running out before the next shipment arrives. The math: reorder point = (sales velocity * lead time) + safety stock

Inputs: velocity (units per day), stdDev (daily demand variation), leadTimeDays, optional serviceLevelPct (85, 90, 95, or 99; default 95), optional manualBufferDays and manualBufferPct.

safety_stock

The cushion that absorbs a bad week. Sized against how much this product's demand actually varies, over the supplier lead time, at a chosen service level. Returns safety stock in units and the equivalent in days of demand.

Inputs: velocity, stdDev, leadTimeDays, optional serviceLevelPct and buffers.

days_of_cover

How long the stock you have plus the stock on order will last at the current sales rate. Higher is better, and this is the single most useful number in inventory planning: a product with 40 units and 8 days of cover is a fire; a product with 40 units and 200 days of cover is a shelf ornament.

Inputs: onHand, optional onOrder, velocity (must be greater than zero).

stockout_risk

When this product runs out, and whether that date lands inside the supplier lead time. Risk is high when the stock runs out before an order placed today could arrive, medium when it runs out inside the lead time plus the review period, and low beyond that. A projected stockout more than two years out comes back as null rather than a date nobody should plan against.

Inputs: onHand, optional onOrder, velocity (must be greater than zero), leadTimeDays, optional reviewPeriodDays (default 7).

economic_order_quantity

The order size where the cost of placing orders and the cost of holding stock balance out: EOQ = sqrt(2 * annual demand * cost per order / holding cost per unit per year) Order smaller than that and the per-order costs add up; order larger and the cash sits on a shelf.

Inputs: annualDemand, orderCost, holdingCostPerUnitPerYear (must be greater than zero).

Brand facts resource

The replinish_facts resource returns a JSON document with the brand name, the spelling disambiguation (Replinish with an i versus the separately published Shopify app named Replenish), the current plan pricing pulled from the same locked constants the rest of the site uses, the full feature list, and a note that Replinish is independent of Shopify.

Read it with:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "resources/read",
  "params": { "uri": "replinish://facts" }
}

The same facts are also published without MCP at https://www.replinish.app/api/facts.json, with Access-Control-Allow-Origin: * and a one-hour cache. Either path returns the same JSON.

Connect from Claude

In Claude Desktop, open Settings, then Connectors, then Add custom connector. Paste the JSON below and save. After Claude restarts, the Replinish tools and the replinish_facts resource are available without further setup.

{
  "mcpServers": {
    "replinish": {
      "url": "https://www.replinish.app/mcp",
      "transport": "streamable-http"
    }
  }
}

Connect from ChatGPT

In ChatGPT, open Settings, then Apps and connectors, then Create a new connector. Set the connector name to Replinish, paste the endpoint URL below, and the tools are available inside any new conversation.

https://www.replinish.app/mcp

Connect from Cursor

In Cursor, open Settings, then MCP, then Add new global MCP server. Paste the JSON below into your ~/.cursor/mcp.json.

{
  "mcpServers": {
    "replinish": {
      "url": "https://www.replinish.app/mcp",
      "transport": "streamable-http"
    }
  }
}

Security and limits

• Read-only. The server has no database connection, no session, and no write paths. There is nothing here a caller can change, intentionally.
• No PII, no store data. The MCP route imports only the pure-math engine in ~/lib/engine/ and the locked constants in ~/components/marketing/seo.ts. It does not import anything from ~/lib/*.server.ts.
• Every argument validated. Tool arguments are checked against the tool's schema before any math runs: present, a real number, finite, and inside the declared range. Nothing reaches the engine unchecked.
• Rate limited. 60 requests per minute per client IP, in an in-memory bucket. Over that, the server returns HTTP 429.
• Body cap. Request bodies over 1 MiB are rejected with HTTP 413. Anything else returns 400 with a JSON-RPC parse error.
• CORS. The dedicated facts endpoint at /api/facts.json serves Access-Control-Allow-Origin: *. The MCP endpoint itself is not CORS-open, because MCP clients are not browser scripts and do not need a CORS handshake.