AI Agent (MCP)

Practical Set runs an MCP (Model Context Protocol) server — a small local program that gives an AI client a fixed set of tools to call, scoped to your own role in the workspace. It works with Claude Desktop, Claude Code, Codex CLI, Cursor, or any other MCP-compatible client, on Team plan and above.

1. 1Settings → API → Create token, and copy it — it's shown once.
2. 2Settings → API → Download the MCP server, and unzip it anywhere.
3. 3Point your client at the unzipped file with the token in its environment — the README in the download has the exact config for Claude Desktop, Claude Code, and Codex CLI.

• Search the catalog by name, manufacturer, model, kind, or category.
• Read one item's full record.
• Create an item — including pricing, a URL, ports, and custom fields.
• Create an item from a product URL — the same thing the app's "Add from URL" does, scraping name/manufacturer/model/price/thumbnail and filling in only what you didn't already specify.
• Re-scrape an existing item's product page to fill in whatever is still blank, without touching fields already set.
• Update any field on an existing item.
• Delete an item (asks for confirmation first).

Try: "Search the library for anything with HDMI in the name" or "Add a Shure SM58 microphone to the library, category Audio, $99."

Try: "Add this to the library: [product URL]" or "That cable is missing a price — pull it from its URL."

• Create a kit (a library item that's a recipe of other items) with a starting parts list.
• Read a kit's recipe, interface ports, and wiring-diagram size.
• Replace a kit's recipe wholesale.
• Replace a kit's interface ports — its inputs/outputs as seen from outside the kit.

Try: "Create a kit called Touchscreen Kiosk with a 24\" touch display, a mini PC, and a wall mount."

• Read a kit's diagram — every node and wire, including its interface boundary nodes.
• Place another library item (or a nested kit) onto the diagram, snapshotting its ports.
• Add, rename, or remove a pin on a node.
• Wire two pins together, or remove a wire.
• Remove a node and everything connected to it.
• Validate the diagram — dangling wires, direction mismatches, orphan nodes, power-budget overruns.

Try: "Open the Touchscreen Kiosk kit's diagram, wire the display's power pin to the mini PC's power output, and validate it."

• List clients (archived ones hidden unless asked for).
• Read one client's full record.
• Create, update, archive/restore, or delete a client.

Try: "Add a new client called Meridian Museums, then archive the old Riverside Client record."

• List projects, filterable by status or client.
• Read one project's full record, including its cost/status rollup.
• Create a project — the org's project and private-project plan limits are enforced the same as the New Project button.
• Update a project's name, code, description, client, status, dates, or budget.
• Change a project's visibility, or delete it.
• Export a project's whole bill of materials as CSV.

Try: "Create a private project called Riverside Install for Meridian Museums, due next month."

• List and read the rows on a project's buying sheet.
• Add a row by snapshotting a library item (with a quantity), or add a fully custom row.
• Patch a row — quantity, status, pricing, dates, vendor, anything else on it.
• Remove a row.

Try: "Add 3 of the Stage Cable from the library to Riverside Install, then mark it Ordered."

Separate from a kit's wiring diagram — this is the implementation-wiring graph for a whole project. A project can have more than one, so the agent lists or creates one by name before working in it.

• List or create a project's Routing graphs.
• Read a graph's nodes and wires.
• Place a BOM row onto the graph, snapshotting its ports.
• Add, rename, or remove a pin; wire two pins together, or remove a wire or a node.
• Validate the graph — the same checks the kit-diagram tools run.

Try: "On Riverside Install's Main Routing graph, place the projector and the media server, and wire the projector's video input to the media server's video output."

The guest/interaction journey, as opposed to the signal wiring — steps, stations, triggers and decisions rather than devices. No ports or pins here; every connection is logical.

• List or create a project's Experience Flow graphs.
• Read a graph's steps and connections.
• Add a step — STEP, STATION, TRIGGER, DECISION, SYSTEM, SOFTWARE, OUTPUT, or LOCATION.
• Rename, re-describe, or move a step; connect two steps, or remove a step or connection.
• Validate the graph for dangling connections and orphan steps.

Try: "On Riverside Install's guest journey, add a Check-in station, then an Interactive Display station after it, and connect them."

• List a project's documents.
• Read one document's full markdown content.
• Create a document (requires the docsCreate plan feature).
• Update a document's title, content, visibility, or parent.
• Delete a document.

Try: "Create a document on Riverside Install called Rack Elevation, and write up the signal chain in it."

• Search publicly published items and kits from every organization.
• Read one published item's full public record.
• Publish a library item or kit — a kit's recipe and diagram publish with it.
• Unpublish a previously published item (existing forks are unaffected).
• Fork a published item into this org's own library as a new, independently editable item.
• Flag a published item for moderator review.

Try: "Search Community for LED panel kits, and fork the one from Bright Labs into our library."