Articles in this section

See more

MCP - Connecting Microsoft Copilot Studio to Softools

Avatar
Mark Dyer
  • Updated
Follow

Overview

Microsoft Copilot Studio can connect AI agents to Softools using Model Context Protocol (MCP), allowing users to ask questions about information held in their Softools Apps.

Once connected, an agent can retrieve and analyse information from published MCP-enabled Views, helping users explore Projects, Risks, Actions and other Records using natural language.

For example, users could ask:

  • Which Projects have the highest priority and need attention first?
  • What are the biggest Risks associated with a particular Project?
  • Which Actions are overdue?
  • Which Projects am I responsible for?

The agent can retrieve information from multiple Softools Apps, subject to each user's existing Softools permissions.

This article explains how to connect Microsoft Copilot Studio to Softools, approve the connection and make the agent available to users through your organisation's chosen channels.

 

Contents

 

Before You Begin

To connect Microsoft Copilot Studio to Softools, you will need:

  • Access to Microsoft Copilot Studio, with permission to configure an agent and its tools.
  • App Builder access to Softools to retrieve the MCP server URL from App Studio.
  • Advanced App Builder permissions in Softools to approve the connection.
  • Softools Apps and Views that have been configured and published for MCP access.

The Copilot Studio configuration and Softools approval can be completed by different people, depending on how your organisation manages its AI assistants.

Before connecting the agent, ensure the relevant Softools Views are available through MCP.

For guidance, see Preparing Softools Apps and Views for MCP.

 

 

Find Your Softools MCP Server URL

Softools provides an MCP server URL for your workspace. This is available to App Builders in App Studio.

To find it:

  1. Open Softools App Studio.
  2. Select Configure from the left-hand menu.
  3. Open MCP connections.
  4. Locate MCP server URL to paste into your AI assistant.
  5. Copy the URL displayed for your workspace.

You will use this URL when configuring the MCP connection in Microsoft Copilot Studio.

The MCP server makes information available through Views that have been enabled and published for MCP. The information users can retrieve also depends on their existing Softools permissions.

 

 

Add the MCP Server in Copilot Studio

Next, configure the AI agent that will connect to Softools.

  1. Open Microsoft Copilot Studio.
  2. Create a new agent or open an existing agent.
  3. Open the agent's Tools configuration.
  4. Select the option to add an MCP server.
  5. Enter a unique connector name, such as Softools MCP.
  6. Paste the MCP server URL copied from Softools App Studio.

**CoPilot01 Screenshot to come

This allows Copilot Studio to discover the MCP tools exposed through your Softools workspace.

 

 

Configure Authentication

Configure the MCP connection to use OAuth authentication.

  1. In the connection's authentication settings, choose OAuth 2.0.
  2. Select Dynamic (with discovery).
  3. Continue with the connection setup.
  4. Select Create new connection in the Copilot Studio sign-in window.

**CoPilot02 Screenshot to come

The sign-in window provides a Softools setup code.

Keep this window open, as you will need the code to approve the connection in Softools.

OAuth authentication allows users to sign in with their own Softools accounts without sharing their passwords with the AI agent.

 

 

Approve the Connection in Softools

An Advanced App Builder must approve the Copilot Studio connection before it can access Softools.

To approve the connection:

  1. Return to Softools App Studio.
  2. Open Configure > MCP connections.
  3. Under Which AI assistant do you want to connect?, select Copilot Studio.
  4. Copy the setup code from the Copilot Studio sign-in window.
  5. Paste it into the Connection setup code Field.
  6. Select Review connection.
  7. Review the connection details and select Approve.

Softools automatically populates the callback URLs required for this connection.

Approving the connection allows Copilot Studio to request access to the workspace. It does not grant unrestricted access to all Softools data.

The Apps, Views and Records available remain subject to the existing Softools permissions and published MCP configuration.

 

 

Complete the Connection in Copilot Studio

Once the connection has been approved in Softools, return to the Copilot Studio sign-in window.

  1. Select Continue to proceed with sign-in.
  2. Sign in using your Softools account.
  3. Review the authorisation request displayed by Softools.
  4. Select Allow read access if the connection is correct.
  5. Return to Copilot Studio and select the authorised connection.

**CoPilot03 Screenshot to come

The Softools authorisation request confirms that the AI assistant is requesting read access to Apps, Views and Records available through the user's existing permissions.

The Softools MCP connection is read-only. It does not provide tools to create, update or delete Softools Records.

Once authorised, the agent can use the Softools MCP tools available through the connection.

 

 

Publish and Share the Agent

After configuring the Softools MCP connection, you can publish the agent and make it available through your organisation's preferred channels.

  1. Return to the agent in Microsoft Copilot Studio.
  2. Review its configuration and available MCP tools.
  3. Select Publish to publish the latest version.
  4. Open Channels and select the channels through which users will access the agent.
  5. Configure the availability and sharing options for your organisation.

**CoPilot04 Screenshot to come

**CoPilot05 Screenshot to come

For example, an agent can be made available through Microsoft 365 Copilot, Microsoft Teams or other supported channels, depending on your organisation's configuration.

Users can then interact with the agent through the interface made available to them.

 

 

User Authentication

When the connection uses End user credential authentication, each person using the agent signs in with their own Softools account.

This ensures that the agent retrieves information using the permissions of the individual user rather than a shared Softools account.

When a user first asks a question requiring Softools data, they may be prompted to connect their account.

The user can then:

  1. Open the connection setup provided by the agent.
  2. Create or select a personal Softools connection.
  3. Sign in using their Softools account.
  4. Review the authorisation request and select Allow read access.
  5. Return to the agent and retry their question if required.

The exact connection interface depends on where the agent is being used.

For a worked example of this process in Microsoft Teams, see Using a Softools AI Assistant in Microsoft Teams.

 

 

Using the Agent with Softools Data

Once connected, users can ask the agent to retrieve and analyse information from published Softools Views.

For example:

Which Projects have the highest priority, and what order should they be reviewed in?

The agent can retrieve information from the Projects View and organise the results using Fields such as Priority, RAG Status and Delivery Date.

Users can also ask questions involving multiple related Apps.

For example:

For New Product Research & Development Facility, what are the biggest Risks and in which order should they be addressed?

Where the Projects and Risks Views are available, the agent can retrieve linked Risk Records, compare their recorded scores and suggest an order of attention.

 

Improving the Quality of Responses

The quality of AI responses depends partly on the information and context provided through published Views.

For example, a Priority Field containing values from Level 1 to Level 5 may need a description explaining which value represents the highest priority.

Adding clear Field descriptions helps the agent interpret the information correctly and provide more reliable recommendations.

For further examples, see Using AI Assistants with Softools Data.

 

 

Managing and Revoking Access

Advanced App Builders can review and manage approved AI assistant connections in Softools.

To manage a connection:

  1. Open Softools App Studio.
  2. Go to Configure > MCP connections.
  3. Locate the relevant connection under Approved AI assistants.
  4. Review the connection details or select Revoke access if the connection is no longer required.

Revoking an approved AI assistant also revokes its associated user grants.

This prevents further access through the revoked connection, but does not necessarily remove information previously retrieved by the AI assistant.

Users may also manage their personal connections through the connection management interface provided by Copilot Studio.

For more information, see Security Considerations When Connecting AI Assistants to Softools.

 

 

Troubleshooting

I cannot find the MCP server URL

Open Softools App Studio > Configure > MCP connections.

The MCP server URL is displayed on this page. Contact an App Builder if you do not have access.

 

Copilot Studio cannot connect to Softools

Check that you have entered the correct MCP server URL and selected OAuth 2.0 > Dynamic (with discovery).

Also confirm that the Copilot Studio connection has been approved in Softools.

 

The connection has not been approved

An Advanced App Builder must complete the approval process in Configure > MCP connections.

Use the setup code generated for the current Copilot Studio connection.

 

A user is asked to set up a connection

This is expected when the agent uses individual user authentication and the user has not yet authorised their Softools account.

Follow the sign-in and authorisation process described above.

 

The connection is successful but the agent cannot retrieve data

Check that the relevant Softools Views have been enabled and published for MCP.

Also confirm that the signed-in user has permission to access the required information.

 

The agent cannot find a particular App or Record

Check the published MCP Views, included Fields and Record permissions.

The agent can only retrieve information made available through the connection.

 

The agent returns incomplete or unexpected information

Review the published View configuration, Field descriptions and any applicable filters.

Ask the agent to explain which Records and Fields it used to produce its response.

AI-generated recommendations should be reviewed against the underlying Softools information before important decisions are made.

 

Related Articles

Related articles

Comments

0 comments

Article is closed for comments.