Connect an Agent

Your assistant is only as useful as what it can see, and today it can't see your register, your spec, or your submittal log, so you copy and paste. Connect it to Submittal Kit over MCP and it works from the same record you do.

What a connected agent can do

• Look up your projects and page through a project's Equipment Register, with each item reporting its datasheet coverage.
• Search your parts and documents, search inside PDFs, and read a document's full extracted text.
• Search the project spec's full text and answer questions from it, citing the sections it drew on.
• Draft, address and send RFIs, including a follow-up that re-asks a sent RFI within its thread, attach files to them, link related RFIs, and record the answer when it comes back.
• Create submittals and prepare them the way you would on the submittal's page: rename it, link its spec sections, add and arrange its sections, choose its equipment and which document prints for each item, run the spec check, and make the O&M Manual that goes with it.
• Read the submittal log, start compiles and check their progress, then mark a submittal as submitted and record the review outcome.
• Share a submittal or a sent RFI with a reviewer by email, turn off a reviewer's link, and record or set aside what a reviewer sends back.
• File what a reviewer sent back, a stamped submittal or an answered RFI, as its response. The PDF is read just as it is when you upload it, and the agent can record the outcome and the day it came back. It fills those in; it won't change an outcome you already recorded.
• Build a submittal's equipment from a BOM you attach or describe, creating any parts your library lacks.
• Queue datasheet discovery for up to 25 parts at once, or hand over datasheets and manuals it found on the web.
• Show you each candidate document and accept or dismiss the ones you approve, without leaving the chat, and unlink a document you change your mind about.
• Read these help guides, so it can answer "how do I" questions about Submittal Kit itself.

What it can never do

A connected agent cannot change billing or company settings. Before anything leaves the building or becomes part of the project's record, it shows you what will happen and waits for your yes in the chat: before it marks a submittal as submitted, before it sends an RFI, and before it emails a reviewer a link. It also asks before it adds a manufacturer you don't have yet, and before it replaces or syncs a submittal's equipment, after showing you what goes. It won't unlink a datasheet whose highlights you reviewed; that stays in the app. And any document it finds for you waits as a candidate until you approve it, on the part's page or in the chat: nothing auto-attaches, so a wrong datasheet costs you a click, not a bad submittal.

Everything it does is a change you could make yourself and can take back the same way: a section it adds can be removed, a reviewer it invites can be turned off, and a submission that was never sent can be withdrawn.

Connect it

1. Point your assistant at your Submittal Kit workspace (most tools ask for a connection URL or pick Submittal Kit from a catalog).
2. It sends you through the standard authorization flow. The consent screen names who is asking, the account and workspace it will act as, and where it redirects. Approve only if you started the connection and recognize the destination. If it names the wrong account, choose Use a different account and sign in with the right one. The connection continues from there.
3. That's it. The agent now acts with your access, and everything it does is attributed to your account.

Each connection can make 120 tool calls a minute. Past that, the agent is told to wait a minute and try again. A request that covers many parts at once, such as datasheet discovery for 25 parts, counts as one call.

Manage connections under Settings, Account, in the Access section: every connected agent is listed with when it connected and when it was last used, and you can revoke one at any time. A revoked agent must reconnect and be approved again.

The help center is agent-readable

Every guide here is also served as plain text at its URL plus .md, and the index lives at /llms.txt. If your assistant needs to know how a feature works, it can read the same pages you do.