Stockly connector

Stockly by Analytos Labs: Stockly, the AI inventory control tower, in Claude and ChatGPT.

What it does

Stockly forecasts demand, sizes each part's Kanban loop (cards, container size, safety stock, reorder point), scores supplier risk, simulates what-if scenarios and keeps a human approval gate on every change. The connector lets you ask about all of it in plain language and, with your confirmation, take the same actions your Stockly account can take in the Stockly app.

Connect

In Claude or ChatGPT, add a connector with the URL https://stockly.analytos.ai/mcp.
Sign in with your own Stockly username and password on the "Connect to Stockly" page.
Ask questions such as "Which SKUs are RED right now?" or "Does a 5-day delay on SKU-1018 matter?"

Every request runs as you, with your Stockly permissions. Disconnecting the connector, or revoking it, ends access straight away.

How actions work

Action tools never change anything on the first call. They return a preview of exactly what would change. Only when called again with confirm=true do they act, and Stockly records the change under your name in its audit log. Submitting purchase orders cannot be undone. Your account's permissions decide which actions are available; without the permission the tool refuses.

Tools (23)

Tool	Type	What it does	Inputs
`approve_recommendation` Approve a recommendation	Action (preview first, then confirm)	Approve a pending Kanban recommendation by rec_id (from list_recommendations; not the sku_id). Stockly sets the card count to the recommended number and adds the buy to the Purchase Queue. confirm: false (default): return a preview and change nothing. true: make the change.	`rec_id`, `notes`, `confirm`
`reject_recommendation` Reject a recommendation	Action (preview first, then confirm)	Reject a pending recommendation by rec_id. reason: at least 10 characters. confirm: false (default): return a preview and change nothing. true: make the change.	`rec_id`, `reason`, `confirm`
`approve_modified` Approve at a different size	Action (preview first, then confirm)	Approve a pending recommendation at a size the planner chose: give exactly one of cards (Kanban card count) or quantity (units), plus a reason. confirm: false (default): return a preview and change nothing. true: make the change.	`rec_id`, `reason`, `cards`, `quantity`, `confirm`
`undo_decision` Undo a decision	Action (preview first, then confirm)	Put an approved or rejected recommendation back to pending, restoring the previous card count. Stockly refuses when the buy was already submitted as a purchase order. confirm: false (default): return a preview and change nothing. true: make the change.	`rec_id`, `reason`, `confirm`
`submit_purchase_orders` Submit purchase orders	Action (preview first, then confirm)	Record pending purchase-queue items (ids from get_purchase_queue) as placed in the ERP under po_number. It does not create the order in the ERP, and once submitted it cannot be undone. confirm: false (default): return a preview and change nothing. true: make the change.	`suggestion_ids`, `po_number`, `confirm`
`apply_resolution` Apply a resolution	Action (preview first, then confirm)	Apply one resolution path (expedite, alternate_supplier, accept_move, raise_buffer) from a scenario run (run_id from run_scenario). The preview is Stockly's own resolution preview. confirm: false (default): return a preview and change nothing. true: make the change.	`run_id`, `sku_id`, `path_key`, `notes`, `confirm`
`set_override` Pin a planning value	Action (preview first, then confirm)	Pin a SKU's reorder point (rop), lot size (lot_size) or safety stock (safety_stock) so the AI stops changing it, optionally until expires_at (YYYY-MM-DD). Replaces an active pin of the same type. confirm: false (default): return a preview and change nothing. true: make the change.	`sku_id`, `override_type`, `value`, `reason`, `expires_at`, `confirm`
`clear_override` Clear a pinned value	Action (preview first, then confirm)	Remove a SKU's pinned rop, lot_size or safety_stock so the AI manages it again. confirm: false (default): return a preview and change nothing. true: make the change.	`sku_id`, `override_type`, `confirm`
`whoami` Who am I	Read	The signed-in Stockly user: name, role, admin flag and the pages and actions the account may use.	-
`get_plant_overview` Plant overview	Read	Plant-wide status in one call: system health, data freshness (the as_of date the data describes), portfolio risk by RAG status with the top-risk SKUs, and the daily AI briefing.	-
`list_recommendations` List recommendations	Read	Stockly's current Kanban recommendation for every SKU: cards now vs recommended (n_current, n_recommended), safety stock, reorder point, working capital, 12-week stockout probability, RAG status, rationale and approval status. Each row's rec_id is the id the approval tools take.	`rag_status`, `status`, `bottleneck_only`, `limit`
`get_sku_detail` SKU detail	Read	Everything about one part (sku_id such as SKU-1001): the full recommendation with SHAP drivers, the demand forecast (P10/P50/P90), the net inventory position and any planner overrides.	`sku_id`
`get_suppliers` Suppliers	Read	Without supplier_id: every supplier, worst first, with RAG status, Supplier Risk Index (SRI), 12-week on-time rate, P80 lead time, drift and fallback flags. With supplier_id (such as SUP-007): that supplier's full scorecard and AI risk assessment.	`supplier_id`
`list_confirmation_exceptions` Vendor confirmation exceptions	Read	Open vendor confirmation problems (DATE_SLIP, SHORT_CONFIRM, PRICE_VARIANCE, ITEM_MISMATCH, NO_CONFIRMATION), worst customer consequence first, with the suggested delay for a delay check.	`severity`, `limit`
`list_orders` Sales or work orders	Read	Open customer sales-order lines (kind='sales', highest value first; filter by customer_id, item_id, priority_tier) or open production work orders (kind='work', earliest start first; filter by item_id, workcenter).	`kind`, `customer_id`, `item_id`, `priority_tier`, `workcenter`, `limit`
`trace_sales_order` Trace a sales order	Read	The chain behind one sales-order line (such as SO-1062-10): customer, finished good, work orders, components and the supplier purchase orders feeding them.	`sales_order_line_id`
`get_production_schedule` Production schedule	Read	Finite-capacity schedule of open work orders against work-centre capacity: summary, late or infeasible jobs, job dates and load. Computed on request; changes nothing.	`planning_date`, `horizon_days`
`get_purchase_queue` Purchase queue	Read	Approved buys grouped by supplier, with each item's id, quantity, reason, dates and status (pending, submitted, reconciled). Item ids are what submit_purchase_orders takes.	`status`
`check_supplier_delay` Check a supplier delay	Read	Deterministic check of whether a supplier delay of delay_days on one SKU matters today: stock, open POs, buffer coverage, affected work and sales orders, and suggested actions. Fast; no simulation.	`sku_id`, `delay_days`
`get_audit_log` Audit log	Read	Who did what and when in Stockly: approvals, rejections, modifications, scenario runs, purchase-order submissions. Filter by sku_id, action (for example APPROVED, REJECTED, MODIFIED, SCENARIO_RUN, PO_SUBMITTED) and dates (YYYY-MM-DD).	`sku_id`, `action`, `date_from`, `date_to`, `limit`
`get_notifications` My notifications	Read	The signed-in user's Stockly notifications, newest first.	`unread_only`, `limit`
`run_scenario` Run a what-if scenario	Simulation	Digital Twin Monte Carlo what-if for one SKU from a plain-English scenario (for example 'SUP-007 delivers 3 weeks late for the next month'). Compares current and recommended settings: weekly stockout risk, working-capital change, verdict and resolution plan. Returns a run_id for preview_resolution. Takes up to two minutes, is recorded in Stockly's audit log as SCENARIO_RUN, and is limited per user per hour.	`sku_id`, `scenario_text`, `n_paths`
`preview_resolution` Preview a resolution	Read	What one resolution path would change after a scenario run: expedite, alternate_supplier, accept_move or raise_buffer. Shows the changes, numbers, residual risk and whether it can be applied. Changes nothing.	`run_id`, `sku_id`, `path_key`

Good to know

rec_id (from list_recommendations) identifies a recommendation; it is not the sku_id.
Stockly's data is loaded from a feed; get_plant_overview reports the as_of date the data describes.
Digital Twin runs take up to two minutes and are limited per user per hour.

Support

Email support@analytos.ai. Website https://stockly.analytos.ai.