Purchase Requisition

MCP server for purchase requisitions you build once and approve per order, with the signed record of each approval. Purchase requisitions you build once and approve per order, with the signed record of each approval.

Works with Claude Desktop, Claude Code, Cursor and any Model Context Protocol client. Runs on your own machine, or hosted with no install.

Product page: https://mcp.zovo.one/s/purchase-requisition — what it does, the tools it exposes, and a live token endpoint.

Install

Hosted, nothing to install. Get a token from https://mcp.zovo.one/mcp/connect (the connect page) or https://mcp.zovo.one/mcp/token (the same token as JSON); a free anonymous one is issued on the spot and a Pro key works the same way. Then point an MCP client at https://mcp.zovo.one/mcp/purchase-requisition over streamable-http and send the token as Authorization: Bearer <token>.

If your client cannot set headers, put the token in the path instead: https://mcp.zovo.one/mcp/purchase-requisition/t/<token>. Both forms work. The bare URL with no token answers 401 on tools/call, so the token is not optional.

Claude Desktop, one click. Download purchase-requisition.mcpb from the latest release and double-click it.

From source. The mirror is self-contained: every @theluckystrike/* dependency is vendored, so a fresh clone builds with no extra setup.

git clone https://github.com/theluckystrike/mcp-purchase-requisition.git
cd mcp-purchase-requisition
npm install && npm run build

Then point your client at the built entry point:

{
  "mcpServers": {
    "purchase-requisition": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-purchase-requisition/dist/index.js"]
    }
  }
}

@theluckystrike/mcp-purchase-requisition is not published on npm yet, so an npx -y @theluckystrike/mcp-purchase-requisition command will fail. The three paths above are the working ones and each is exercised by CI.

Read-only mirror of mcp-servers/servers/purchase-requisition. See MIRROR.md.

theluckystrike/mcp-purchase-requisition MCP server

In the official MCP Registry (io.github.theluckystrike/purchase-requisition). Checklists you build once and run many times, and the dated record of each run that somebody signs. A purchase-requisition is a named list of steps, optionally grouped into sections, each one required or optional. A run is one pass of that purchase-requisition against a job: every step is marked pass, fail or not applicable, with who marked it and on what day, and a note that says what was found. run_sign_off then puts a name and a date on it and freezes it.

The one rule that decides everything else

If somebody edits the purchase-requisition afterwards, adds a step or deletes one, every run already in progress keeps the list it started with, and the version it was copied from is recorded on the run.

That is not a caching convenience. A purchase-requisition somebody ticked and signed has to be the list they actually saw. A run that read its steps live from the purchase-requisition would mean a signed handover certificate for eleven checks when the person signing it saw ten, with no field in the record showing that it had happened. It also means deleting a purchase-requisition leaves its runs readable and complete, which is what you want the year afterwards when somebody asks what was checked.

Two smaller rules follow from it:

• Nothing derived is stored. The pass, fail and outstanding counts, the percentage, and whether a run can be signed off are worked out on every call from the run's own steps. A stored "complete" flag is a fact about the afternoon somebody last looked, and complete here is a reading: it appears when the last step is answered and goes away again when one is put back to pending.
• Not applicable is not a pass. na counts as ANSWERED and never as passed. A step that was looked at and dismissed is a different fact from a step that passed, and merging the two is how a purchase-requisition reports full marks for a job where half the steps did not apply.

What blocks a signature

A required step that is unanswered, a required step that failed, an unanswered optional step, or a run with no steps. force: true signs anyway, and the exceptions stay on the record and print on the report under "Signed with exceptions". They are not lost, and they are not silent.

Install

One-click (.mcpb): download purchase-requisition.mcpb from the latest release and double-click it in Claude Desktop: https://github.com/theluckystrike/mcp-servers/releases/latest

npm publish for @theluckystrike/mcp-purchase-requisition is pending, so the npx line below returns 404 today. Build from source in the meantime; see llms-install.md.

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "purchase-requisition": {
      "command": "npx",
      "args": ["-y", "@theluckystrike/mcp-purchase-requisition"]
    }
  }
}

Claude Code

claude mcp add purchase-requisition -- npx -y @theluckystrike/mcp-purchase-requisition

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (project), same entry as Claude Desktop.

Tools

Tool	What it does
purchase-requisition_create	Create a reusable purchase-requisition: a name, a category, a description
purchase-requisition_item_add	Add a step: the text, a section heading, and whether it is required
purchase-requisition_item_remove	Remove a step and bump the version. Runs already under way keep it
purchase-requisition_show	One purchase-requisition, grouped by section, with a blank printable copy on request
purchase-requisition_list	Every purchase-requisition with its version, step count and how many runs came from it
purchase-requisition_delete	Delete a purchase-requisition. Its runs stay readable, because each carries its own copy
run_start	Start a dated run against a job. The steps are copied into it at this point
run_check	Mark one step pass, fail or na, with who and when and what was found
run_show	The run: every step with its answer, the counts, the failures, and what blocks sign-off
run_list	Runs newest first, filtered by purchase-requisition, status, reference, or only those with failures
run_sign_off	Sign off with a name and a date, which freezes the run
run_status	Reopen a complete run, or abandon one when the job did not happen
run_report	The run as text on every tier. Pro also writes it to out_path as a .txt file
run_delete	Delete a run. A signed-off one is refused
license_status	Which tier this install is on and where the key came from
license_activate	Store a Pro key for this server

There is also a resource, purchase-requisition://contract, carrying the snapshot rule, the item states, the run status machine, what blocks a sign-off and where this server writes; and a prompt, run_the_purchase-requisition, that walks the whole job in order.

Free vs Pro

	Free	Pro
Checklists you keep	3	unlimited
Runs of them	unlimited	unlimited
Steps per purchase-requisition	up to 500	up to 500
run_show, run_list, the counts	yes	yes
The run report as text	yes	yes
Writing the report to a file with out_path	no	yes

The meter is on how many DIFFERENT purchase-requisitions you keep, not on how many jobs you check. A trade with one pre-delivery check, one handover sheet and one snag list runs its whole year inside the free tier. Runs are never capped, because capping the running of a purchase-requisition would cap the only thing a purchase-requisition is for. Deleting a purchase-requisition frees a slot.

Get Pro: https://mcp.zovo.one/buy/purchase-requisition (one-time), or all servers for one price at https://mcp.zovo.one/buy/bundle

Privacy

All data stays local, in ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/purchase-requisition/. There is no network call anywhere in this server, no API key, and no account. The only file it reads that it does not own is the shared business profile, for the name and address at the top of a printed report, and it never writes to it.

Built by theluckystrike. Support: [email protected]