Skip to content

ERP and MRP your AI agents can use

Letting an agent loose on stock, orders and safety records is only a good idea if it can't do more than the person it works for. Knotworking exposes its modules to agents over MCP with exactly that rule.

Every module is an MCP tool set

More than 200 tools over MCP cover the modules, from listing tasks to recording CCP readings. They work with Claude Code, Claude Desktop and any client that speaks MCP. In the app, an assistant sidebar drives each person's own Claude Code, on their machine and subscription, with their permissions.

What an agent can and can't do

Agents sign in with a person's API token, and the token can only narrow that person's access.

Read-only by default
A new token can only read. Writing is a box you tick, per token.
Limited by module
Untick a module and its tools disappear from the agent's list.
Always expires
After 30, 90, 180 or 365 days; 90 unless you choose. There is no never-expires option.
Stored as a hash
A token is shown once and revoked from the page you made it on.
Rate limited
120 tool calls a minute per token by default.
Some actions need a person
Approvals, sign-offs and safety records need a signed-in person.

What it is good for

Ask which products are short for next week's schedule, have an agent prepare purchase orders for a person to review, or summarise open corrective actions before a review. The agent sees what you would see, and nothing else.

The modules involved

Each is a separate module, switched on in the installation and then granted person by person. How modules work.

Assistant
A ⌘J sidebar that drives each person's own Claude Code, on their machine and subscription, with their permissions.
Tasks and projects
Team workflows, priorities, labels, estimates, subtasks and dependencies, in lists and boards. Projects carry status, health, a lead and live progress.
Stock
Locations, dated lots, goods-in and an append-only movement ledger with traceability.
Production
A batch schedule from approved recipes, material requirements against stock, and batch records that consume and produce lots.
Purchasing
Purchase orders from the shortfall, goods-in against an order, three-way invoice matching and supplier performance.
Quality
Complaints, CAPA, internal audits, mock recalls, incidents and authenticity assessments.

Questions

Which agents work with it?
Claude Code, Claude Desktop and any client that speaks MCP.
Can an agent see data I can't?
No. A token has its owner's access at most, and can be narrowed to a list of modules.
Can an agent approve things?
No. Approvals, sign-offs and safety records need a signed-in person.
Does my data go to a third party?
Knotworking itself runs on your server. What an agent does with data it reads depends on the client you connect, so choose that with care.