Back to Documentation

AI workflows: the MCP Server in practice

With Sutram connected to an AI assistant (such as Claude) through the MCP Server, you talk to your project's content: finding, reading, commenting on, correcting, and versioning documents in natural language, without leaving the chat. This tutorial walks through a complete cycle, end to end.

Prerequisites: the project must be connected to the assistant (see Connecting Sutram to Claude), and MCP access requires the Pro or higher plan. For the reference of each tool mentioned, see the Sutram MCP Server Guide.

The cycle you'll go through

Get oriented  →  Find  →  Read and assess  →  Comment  →  Correct and version  →  Close the loop

The core idea: you describe your intent in plain language; the assistant chooses and chains the right tools. You don't need to know the tool names — but it helps to understand what happens behind the scenes, and that's what this guide shows.

Step 1: Get oriented in the project

Start with an overview:

"Show me an overview of my Sutram project and what's in it."

Behind the scenes, the assistant queries the project information and lists the root, returning a structured summary of the folders and items. It's the starting point for knowing where things are.

Step 2: Find what you need

Ask by whichever path makes sense — by metadata or by subject:

"Find the documents tagged with category = contract and year = 2026." "What do we have about the refund policy? Search the wiki and tell me the sources."

The assistant runs the tag search or the full-text search on the Wiki and brings back the results with each item's path. (See Content search and discovery.)

Step 3: Read and assess

Point to a document for the assistant to open and analyze:

"Open the ACME contract and summarize the main points; flag any outdated clause."

The assistant downloads the file content, reads it, and responds — including flagging any issues you asked it to watch for.

Step 4: Leave feedback with comments

Record observations in the document itself, for the team to see:

"Comment on this file: section 3 references a repealed regulation."

The assistant creates a comment — on the whole file or anchored to a passage of a markdown document. The comment appears in real time in the web interface, exactly as if you had written it by hand. This is how AI takes part in the review alongside the team.

Step 5: Correct and version

Here's the heart of the workflow. Ask for the correction:

"Enable versioning on this document and fix the section 3 reference to the current regulation."

Behind the scenes, the assistant runs the versioning cycle:

  1. Enables versioning (if the file is still a reference)
  2. Performs a check-out (locks it for exclusive editing)
  3. Edits and uploads the new version of the file
  4. Performs a check-in — the file becomes a draft

The draft is not published automatically: you review it first. Once you approve:

"Looks great, go ahead and publish."

...the assistant publishes the version, creating a permanent snapshot in the history. For documents under governance (Document Classes), the path is to create a new revision and move it through the lifecycle with transitions (see Governance in practice: states and transitions).

Safety: publishing and other sensitive actions ask for your confirmation; destructive tools (delete) always confirm before acting. The draft exists precisely so you can review before making something official.

Step 6: Close the review loop

Collaboration goes both ways. When you or colleagues comment on the documents, the assistant keeps up:

"Look at the open comments on this doc and reply to the ones that have already been resolved."

It lists the comments, replies to the threads, and marks as resolved what has been addressed — keeping the review organized.

Best practices when working with AI

  • Be explicit — say "in Sutram" when the request could be confused with a local file; or apply the Sutram Assistant Preset (see Connecting Sutram to Claude) so the assistant assumes that context from the start.
  • Review before publishing — the draft → published cycle exists for this; let the assistant stop at the draft and check.
  • Confirm external actions — publishing and posting to the chat are visible to the team; confirm before doing so.
  • Take advantage of batching — AI shines at repetitive tasks: creating dozens of records, tagging in bulk, reorganizing folders.

An end-to-end example

You:  Overview of my Sutram project.
AI:   [lists folders and items at the root] You have Contracts, Reports, and Documentation.

You:  In Contracts, find the ones from 2026 and tell me if any cites regulation X.
AI:   [tag search year=2026] Found 3; the "ACME Contract" cites regulation X (repealed).

You:  Comment on that in the file and correct it to the current regulation.
AI:   [creates comment] [enables versioning → check-out → edits → check-in]
      Comment recorded and correction made — it's in draft for your review.

You:  Perfect, publish it.
AI:   [publishes the version] Published as a new version. Loop closed.

Frequently Asked Questions

Q: Do I need to know the MCP tool names?

A: No. You talk in natural language and the assistant chooses the tools. Knowing what they do helps you ask precisely and understand what happened — that's what the Sutram MCP Server Guide is for.

Q: Can the AI publish or delete without me seeing it?

A: Sensitive actions ask for confirmation, and destructive ones always confirm. The default is for the AI to stop at the draft; publishing happens when you approve.

Q: Does this work in any client?

A: It works in MCP-compatible clients — via OAuth connector (claude.ai, Desktop) or key bridge (Claude Code, Cursor). See Connecting Sutram to Claude.

Next Steps

  • Connecting Sutram to Claude — to get the assistant ready
  • Sutram MCP Server Guide — the reference for all the tools
  • File Versioning System — what happens during check-out/check-in/publish

Document Version: 1.0 Last Updated: July 2026 Author: Sutram Development Team

Back to Documentation