Hands-on Lab (WIP)

Before you begin Start here

0.1 · The Scenario

0.2 · Learning Objectives

0.3 · How to Use This Handbook

Self-paced and self-contained - follow it in a facilitated session or work through it independently.

0.4 · Prerequisites - set up before the session

Get every developer's machine ready before Stage 1 so the 15-minute CLI slot is spent on the workflow, not on installs. Everything here uses the demo environment and synthetic data only, no production accounts or real vendor data.

Tick each section as you complete it. Once all four are checked the lab stages unlock automatically.

0.4.1 · Tooling on each machine

1. Check if Node.js + npm are already installed - open your terminal and run:
   Copynode -v && npm -v

   If you see two version numbers (e.g. v20.x.x and 10.x.x), Node and npm are installed - skip to step 2. If you get "command not found", install Node.js LTS from the terminal:
   
   Mac: brew install node
   Windows: winget install OpenJS.NodeJS.LTS
   
   Then re-run the check before continuing.

2. Docusign Agreement CLI installed globally:
   Copynpm i -g @docusign/agreement-cli@1.1.0-beta
3. Confirm the docusign command is on the PATH, running it with no arguments prints the help / command list:
   Copydocusign

0.4.2 · Accounts & access

• A Docusign developer / demo account with Docusign CLI (open beta) access enabled.
• Permissions to configure Agreement Manager (agreement types, fields, ingestion) in that account.
• For Stage 3: a Microsoft 365 / Copilot Studio environment where you can create agents and add Model Context Protocol tools. We have provisioned lab accounts for this session - use the number on the paper provided by a volunteer and replace X with your assigned number:
  • account:
  • password:
  • use your existing environment, or provision a trial environment via Trial Environment Setup, Steps 1–4:
    https://microsoft.github.io/agent-academy/recruit/00-course-setup/

0.4.3 · Download the workshop resources

The workshop resource bundle ships the Pharma/Medtech procurement manifest, training documents, and ingest contracts ready to use. All data is synthetic.

Download the ZIP and unzip it into the root of your working directory:

• Manifest: Bengaluru-meetup/workshop-resources/agreement-manager-manifest.json
• Training docs (7 sample agreements): Bengaluru-meetup/workshop-resources/files/train/
• Ingest contracts (4 contracts): Bengaluru-meetup/workshop-resources/files/ingest/

0.4.4 · Import the Workflow

The workshop kit includes a pre-built Workflow Builder workflow for the Stage 3 renewal exercise. Import it now so it is ready before you reach Stage 3.

1. In your Docusign demo account (apps-d.docusign.com), go to the Agreements tab and click Workflows in the left navigation.
2. Click Import and upload Fontara Renewal Order Form.zip from workshop-resources/workflow/ in the workshop kit.
3. Click Save, then Review, then Publish.

Stage 1 · Docusign CLI + Agreement CLI 15 min

This is where you start Fontara's build. You log in with the Docusign Agreement CLI, define Fontara's procurement agreement types & fields as code, then bulk-ingest their sample repository. The pattern to emphasize: export legacy repository → standardize via Docusign CLI → ingest via API at scale.

1.1 · Authenticate

Browser-based OAuth (PKCE). Each developer logs in with their own Docusign developer account - no shared or production credentials. On a headless machine add --no-browser to get a printed URL.

Expect: a browser tab opens, you grant access, CLI prints the authenticated account ID. Then verify:

Expect: Authentication is valid - confirms the token is live before continuing.

1.2 · Scaffold the workspace & project

Before scaffolding, create a dedicated working folder so all CLI output lands in one place. Pick either option below - you only need one.

Option A - Terminal (recommended):

Option B - Manually:

• Create a new folder anywhere on your machine (e.g. docusign-workshop on your Desktop).
• Open a terminal window and cd into it:
  cd ~/Desktop/docusign-workshop

Now scaffold the project structure. Use lowercase agreement-manager (PascalCase breaks the directory layout in v1.0.7+).

Creates: demo-workspace/demo-project/agreement-manager/ with configs/, files/train/, files/test/, tests/, and more.

1.3 · Drop in the Pharma/Medtech procurement manifest

The workshop resources ship a pre-built manifest - 3 custom agreement types (Clinical Trial Supply, CRO Services, Medical Device Supply) and 9 custom pharma fields.

Copy the manifest into your workspace configs folder:

Copy the 7 training documents (sample agreements the AI learns from) into the workspace train folder:

Pull your account's current catalog - agm upload needs this to know which standard fields and agreement types already exist before it can add your custom ones:

Creates: standard-catalog.json (Docusign's built-in types/fields) and custom-catalog.json (any existing customizations in your account). The upload step reads both to avoid conflicts.

Upload the manifest - creates all fields, creates all agreement types, maps fields to types, uploads training docs, and triggers AI training in one command:

Expect: CLI walks through each step and confirms fields created, types created, training uploaded. AI training runs asynchronously - extraction results appear in the UI after a few minutes.

1.5 · Test the customizations Informational

Generate a test template pre-filled with what Iris actually extracted, then compare against known ground-truth values to validate extraction accuracy before promoting to production.

Edit the generated template to set expected (ground-truth) values, then score accuracy:

Expect: a pass/fail comparison of extracted vs. expected values per field. Use this to tune your manifest before promoting to production.

1.4 · Bulk-ingest the Pharma/Medtech contract set

Ingest 4 real-world-style pharma contracts at once - the "entire legacy repository, not one file at a time" moment. Go back to the repo root first.

Preview what will be uploaded without ingesting:

When ready, ingest for real:

Expect: CLI confirms 4 files uploaded. Documents appear in Agreement Manager within a few minutes as Navigator indexes them asynchronously.

1.6 · Deploy to production Informational

The same validated config promotes demo → production (or replicates across client accounts) - just re-auth and re-upload. The manifest is the portable asset you hand to each client.

Stage 2 · Extraction surfaces in Agreement Manager

Now Fontara's agreements stop being flat documents. Once the corpus is ingested, Docusign's AI (Iris) reads each one and populates the fields you defined in Stage 1. This is where the "static file → structured data" moment lands for the room - procurement sees answers instead of files.

2.1 · Open Agreement Manager and confirm the corpus

1. Go to apps-d.docusign.com → Agreements → Completed.
2. Confirm all 4 ingested contracts appear with their agreement type auto-applied:
   
   • 3 × Clinical Trial Supply Agreement
   • 2 × CRO Services Agreement
   • 2 × Medical Device Supply Agreement
3. Open one document and show the auto-extracted fields panel side-by-side with the source PDF - this is the "Iris read the contract" moment.

Expect: each document lands in the correct custom type (not Miscellaneous or generic). If a document lands in the wrong type, the manifest's aiDefinition is the tuning lever.

2.2 · Pharma/Medtech fields to highlight on screen

These are the 9 custom fields defined in the manifest. Walk through each type and show at least one extracted value per group.

2.3 · Standard fields that also fire on these types

In addition to the 9 custom fields above, the manifest enables standard fields for each type. These surface automatically alongside the custom extractions.

Talking point: procurement now has payment terms, governing law, termination notice, and renewal dates extracted automatically - without any manual review - across the entire corpus. This is what feeds the AI agent in Stage 3.

Stage 3 · MCP via Microsoft Copilot Studio 30 min

Now the payoff for Fontara's procurement team: they stop hunting through files and just ask. You'll connect the Docusign MCP server to Copilot Studio and query Fontara's corpus in plain English, including agreements with vendors like MarketPulse Dynamics and Momentum Driver from the sample set.

3.1 · Create & configure the agent

1. Sign in at copilotstudio.microsoft.com.
2. Create a blank agent and name it "Procurement Agent".
3. Enable generative orchestration: Settings → Generative AI → set to "Yes".
4. Add an agent description, e.g.:

5. Add system instructions: click Edit on Instructions and paste the block below - these ensure the agent always calls Docusign tools for live data and never fabricates agreement state.

3.2 · Add the Docusign MCP Demo connector

1. Open the Tools tab → + Add a tool → Model Context Protocol.
2. Search for and select "Docusign MCP Demo".
3. Create a new connection - when prompted for authentication, select OAuth 2.0 and sign in with your Docusign demo account credentials (apps-d.docusign.com). Allow the requested access. Do not use NoAuth or API key - the connector requires OAuth.
4. Add and configure the tool, then review the available tools (all enabled by default).
5. Publish the agent.

3.3 · Verify the connection

Each prompt should trigger a tools/call in the activity map.

3.4 · Scenario A - Vendor Agreement Insights

Run these prompts in order

Scenario A demonstrates: portfolio-wide renewal visibility · Agreement Manager + Iris extracting parties, dates, payment terms, and renewal notice periods · off-standard terms detection · vendor-to-vendor compliance contrast.

3.5 · Scenario B - Workflow Builder Orchestration

Run these prompts in order

Scenario B demonstrates: agent calling getWorkflowTriggerRequirements before triggering · Workflow Builder routing logic · human-in-the-loop approval for high-value renewals · Agreement Prep generating the order form for vendor signature.

3.6 · Scenario C - Track Workflow & Agreement Status

Run these prompts in order

Scenario C demonstrates: in-flight workflow monitoring · eSign envelope status check · nudging stalled vendor signers without leaving chat · expiry and escalation awareness.

3.7 · Publish to Microsoft Teams & M365 Copilot

Always test in Copilot Studio before publishing to channels. Use the Test your agent pane with the activity map turned on for every iteration. Verify each prompt produces a tools/call against the Docusign MCP Demo connector and that the response is grounded. Only then publish.

1. Open the Channels tab in your agent.
2. Select Microsoft 365 and Microsoft Teams. Tick Make agent available in Microsoft 365 Copilot, then click Add channel.
3. Click Availability Options → Show to everyone in my org.
4. From Agent Preview, edit icon, short and long description, then click Save.
5. Click Publish (top-right) and confirm. Wait for publish to complete.
6. From the channel panel, click See Agent in Teams or See Agent in Microsoft 365 and test from that surface.

3.8 · Sample prompt library

A portable library of prompts to extend the demo beyond the three core scenarios. Replace anything in [BRACKETS] with data from your demo account - vendor names, emails, and contract values.

3.9 · Troubleshooting

Stage 4 · Agent Studio - Procurement Agent 10 min

Agent Studio is Docusign's native agent builder - an alternative path to Stage 3's Copilot Studio approach. Instead of connecting an external AI platform via MCP, you build the Procurement agent directly inside Docusign, grounded on the same Agreement Manager corpus. The result: Fontara's team gets contract intelligence without leaving the Docusign platform.

4.1 · Create the Procurement Agent

1. In your Docusign developer account, navigate to Agent Studio.
2. Click Create a draft agent.
3. In the description box, paste the following and click Create Agent:

4.2 · Name & configure the agent

1. Set the agent name to Fontara Procurement Agent.
2. Under Define what starts this agent, confirm both In chat and In a workflow are enabled.
3. Paste the instructions below into the Instructions field:

4.4 · Test the agent

Click Test (top right) and run these prompts in order. Each should return grounded data from your Agreement Manager corpus.