Articles in this section

Connecting AI Assistants to Softools Using MCP

Avatar
Mark Dyer
  • Updated
Follow

Overview

Softools can connect to supported AI assistants using the Model Context Protocol (MCP). This allows users to ask questions about information held in Softools Apps using natural language, while Softools continues to control which data is available and which Records each user can access.

For example, a user could ask an AI assistant to:

  • Show Projects grouped by Status and ordered by Delivery Date.
  • Find Red or Amber Projects where they are the Project Lead.
  • Summarise information returned from a Softools View.
  • Search configured Data Lens Fields for references to risks, delays or other topics.

Once connected, users can explore information across multiple Softools Apps through a conversation, without needing to create a new Report for every question.

The MCP tools covered in this article provide read-only access. They do not allow the AI assistant to create, update or delete Softools Records.

Softools provides three connection routes:

  • Claude – using Claude's published identity and detected sign-in settings.
  • Microsoft Copilot Studio – using the Copilot Studio setup-code flow.
  • Manual / another AI assistant – for other compatible MCP clients. This article uses ChatGPT as a worked example.

 

 

Contents

 

One Connection for All Apps, Views and Users

An Advanced App Builder approves an AI assistant once at Softools site level. The approval is not tied to one App, one View or one user.

After the assistant has been approved, any published MCP-enabled View across the site can be used through the same assistant connection. If additional Views are exposed later, users do not need to set up a new connection for each View or App.

All authorised users can use the same approved AI assistant connection, but each signs in with their own Softools account and grants read access. They do not share login details or access rights. Softools applies each signed-in user's existing App, View and Record permissions.

For example, an App Builder could approve Claude and expose a Projects View. Later, the builder could expose Views from the Actions and Risks Apps. Those Views would be available through the existing Claude connection to users who have permission to access them.

Connection approval and user authorisation are different steps. The App Builder approves the AI assistant once for the site. Each user then authorises that assistant to access the information available to their own Softools account.

 

Preparing a View for MCP

An AI assistant accesses Softools data through published Views that have been exposed as MCP tools.

If a suitable Table Report already exists, the quickest way to prepare a View is to create it directly from that report.

  1. Open the App in App Studio.
  2. Go to Reports > Lists / Tables.
  3. Open the Table Report containing the Fields you want to make available.
  4. Select the dropdown beside Save.
  5. Select Create a View from this Table.

Enter a View Title and View Identifier, then select Create.

The current Table Report configuration is used as the starting point for the new View.

For further guidance on creating Views and configuring their Fields, see Creating and Configuring Views.

Once the View is ready, it can be exposed as an MCP tool using the steps below.

 

 

Exposing the View through MCP

Open the View and expand API Access.

  1. Enable Expose this published View as an MCP tool.
  2. Add a useful description for the View.
  3. Add descriptions to the Fields exposed by the View where possible.
  4. Save the View.
  5. Publish the App.

Descriptions help the AI assistant understand what the View and its Fields represent. Where a Field description is not available, Softools can use the Field label and type as a fallback.

Important: Publishing the App activates changes to the MCP tool. A saved but unpublished change is not available to connected AI assistants.

 
 

Enabling Data Lens (Optional)

A standard MCP View tool can query the structured information exposed by the View. Data Lens adds a semantic search tool, allowing an AI assistant to search configured Fields based on meaning rather than requiring an exact value match. The semantic search returns a maximum of 40 ranked matches and isn't designed for retrieving a complete dataset.

To enable Data Lens:

  1. Open the App in App Studio.
  2. Go to App Configuration > Advanced Config.
  3. Expand Data Lens.
  4. Enable Data Lens.
  5. Select the Fields that should be searchable.
  6. Save and publish the App.

Up to 10 Data Lens Fields can be selected for an App.

If required, enable Data Lens Translation to support non-English search text and queries.

When Data Lens is enabled, an MCP-enabled View can expose two tools:

  • Query [View] – queries the Records and Fields exposed by the View.
  • Search [View] (Data Lens) – performs semantic searches using the configured Data Lens Fields.

Search results still contain only information exposed by the relevant View.

 
 

Approving an AI Assistant

Softools App Builder

Before end users can connect an AI assistant, an Advanced App Builder must approve the assistant for the Softools site.

Go to Configure > MCP connections.

Softools provides three connection routes:

  • Claude
  • Copilot Studio
  • Manual / another AI assistant

Approval is performed once at site level for each AI assistant, not separately for each App, View or user. Newly published MCP-enabled Views can be used through the existing approved connection.

Each user signs in with their own Softools account and authorises read access. Approving an assistant does not give users access to Apps, Views or Records they could not otherwise access.

 
 

Connecting Claude

Approve Claude in Softools

Softools App Builder

  1. Go to Configure > MCP connections.
  2. Select Claude.
  3. Check the detected client details.
  4. Select Approve Claude.

Softools provides the MCP server URL that users will add to Claude.

No client secret is required for this connection method.

 

Add Softools to Claude

Each user connects Claude to Softools using their own Softools account. The AI assistant only has access to information that the signed-in user is permitted to view.

In Claude:

  1. Go to Customize > Connectors.
  2. Select Add > Add custom connector.
  3. Enter a name such as Softools.
  4. Paste the MCP server URL supplied by Softools.
  5. Select Continue.

For example:

https://api-gateway.softools.net/mcp/demo

Use the URL shown in your own Softools site.

Claude should detect the Softools OAuth settings.

Keep:

  • Authentication: Sign in now
  • OAuth client: Use Claude's published identity

Then select Add.

 

Sign In and Allow Read Access

Claude opens the Softools authentication flow.

If you are not already signed in to Softools, sign in using your normal Softools account.

Review the site and requested access, then select Allow read access.

If Softools does not return you to Claude automatically, select Continue on the Returning to your AI assistant screen.

 

Configure Claude Tool Permissions

Once connected, Claude displays the Softools read-only tools available to you.

These can include:

  • Current Softools user
  • Query [App] – [View]
  • Search [App] – [View] (Data Lens), where Data Lens is enabled

Claude allows you to control whether each tool can run automatically, requires approval, or is blocked.

For example, you might allow the standard Query tool automatically but require approval for each Data Lens search.

Note: Changing Claude's approval setting does not change your Softools permissions. It only controls whether Claude asks before using the tool.

 
 

Connecting Microsoft Copilot Studio

Softools App Builder and Copilot Studio Administrator

Softools also supports connecting Microsoft Copilot Studio through MCP.

The Copilot Studio flow uses a setup code generated during the Microsoft connection process.

 

In Copilot Studio

  1. Add an MCP server using the MCP server URL provided by Softools.
  2. Give the connector a unique name.
  3. Select OAuth 2.0 > Dynamic (with discovery).
  4. Select Create new connection.

The Copilot sign-in window displays a Softools setup code.

 

In Softools

  1. Go to Configure > MCP connections.
  2. Select Copilot Studio.
  3. Paste the code into Connection setup code.
  4. Select Review connection.
  5. Review and approve the connection. Softools fills in the callback URLs automatically.

 

Back in Copilot Studio

Return to the sign-in window and select Continue.

Sign in using your own Softools account, accept read access, then select the connection in Copilot Studio.

 

Note: Microsoft environments and Copilot Studio licensing can vary, so the exact Microsoft screens may differ. The Softools steps above reflect the connection flow provided by Softools.

 
 

Connecting Another AI Assistant Manually

Softools can also connect to other MCP-compatible AI assistants through Manual / another AI assistant.

The exact setup depends on the registration methods supported by the chosen assistant.

The following example shows a connection to ChatGPT using a Manual public client.

 

Example: Connecting ChatGPT

Approve ChatGPT in Softools

  1. Go to Configure > MCP connections.
  2. Select Manual / another AI assistant.
  3. Enter a Client name, such as ChatGPT.
  4. Select Manual public client.
  5. Enter the exact callback URL supplied by ChatGPT.
  6. Select Approve client.

For the ChatGPT configuration used in this example, the callback URL is:

https://chatgpt.com/connector_platform_oauth_redirect

Softools creates a Public Client ID.

The public client ID identifies the approved OAuth client. It is not a password or client secret and can be supplied to users who need to configure the approved connection.

Important: Only approve callback URLs supplied by an AI assistant you trust.

 

Add the Softools MCP Server in ChatGPT

Softools End Users

In ChatGPT:

  1. Open Plugins.
  2. Select Add > Add custom MCP server.
  3. Enter a name and optional description.
  4. Select Server URL.
  5. Enter the Softools MCP server URL: https://api-gateway.softools.net/mcp/demo.
  6. Select OAuth authentication.
  7. Open Advanced OAuth settings.
  8. Select User-Defined OAuth Client.
  9. Paste the Public Client ID generated by Softools.
  10. Leave the OAuth client secret blank.
  11. Keep the token endpoint authentication method set to none.

The Softools OAuth endpoints and scopes should be discovered automatically from the MCP server.

Note: The callback URL and OAuth settings shown are specific to the ChatGPT connection used in this example. Other AI assistants may require different settings, and ChatGPT's configuration options may change over time. Always use the callback URL provided by the AI assistant you are connecting.

Create the plugin and continue to Softools.

Sign in with your own Softools account and select Allow read access.

Once authorised, the Softools tools become available to ChatGPT.

 

Asking Questions About Softools Data

After the connection is complete, users can ask natural-language questions about the Records available through the exposed Views.

For example:

Show me all Projects in the Softools Project App, including their Status, Priority and Delivery Date. Group them by Status, with Red first, followed by Amber and Green, and sort each group by Delivery Date, earliest first.

Claude can retrieve the Records, group them into Red, Amber and Green sections, and sort each group by the requested Delivery Date.

The same View can also be queried from another connected assistant. In this example, ChatGPT uses the manually configured MCP connection to query and analyse the Project data.

The AI assistant can present the same Softools information in different ways without requiring an additional report to be created for every question.

Important: AI-generated summaries and observations should be checked against the underlying Softools Records where accuracy is important.

 
 

Using the Current Softools User

Softools exposes a Current Softools user tool alongside the View tools.

This allows an AI assistant to identify the user who has authenticated the connection and use that information when querying Person Fields.

For example:

Show me the Projects where I am the Project Lead that are Status Red or Amber.

The user does not need to include their own name in the prompt.

This can be useful for questions such as:

  • What Projects am I responsible for?
  • Show my open Actions.
  • Which Red or Amber Projects am I leading?
  • What Records currently reference me?

The answer remains limited by the signed-in user's existing Softools permissions.

 
 

Searching with Data Lens

Where Data Lens is enabled, the AI assistant can use the separate Data Lens Search tool for semantic searches.

For example:

Search my Softools Projects for any that mention delivery delays, resource shortages or risks to completion. Summarise the issues identified and which Projects may need attention.

If the tool is configured to require approval, the AI assistant prompts the user before performing the search.

The assistant then returns the most relevant Records and information available through the View.

The quality of the search depends on the information stored in the selected Data Lens Fields and the Fields exposed through the View.

If the relevant descriptive information is not available through the View, the AI assistant may be able to identify potentially relevant Records without being able to explain the underlying issue in detail.

 
 

Managing Approved AI Assistants

Softools App Builder

Go to Configure > MCP connections to review the AI assistants approved for the site.

The Approved AI assistants section shows the registered clients and provides a Revoke access option.

Revoking an AI assistant also revokes its associated user grants. If the assistant needs to be used again, it must be approved again.

Individual users can also disconnect the Softools connection from within their AI assistant.

 

Troubleshooting

The View Does Not Appear as an AI Tool

Check that:

  • Expose this published View as an MCP tool is enabled.
  • The App has been published since the change was made.
  • The user has permission to access the relevant information.

 

The AI Assistant Cannot Connect

Check that:

  • The correct site MCP server URL is being used.
  • The AI assistant has been approved under Configure > MCP connections.
  • The configured authentication method matches the connection route being used.

 

OAuth Authentication Fails

For a manual client, check that:

  • The exact callback URL matches the URL supplied by the AI assistant.
  • The correct Public Client ID is being used.
  • No client secret has been added for a public client.
  • The assistant supports the authentication method required by Softools.

 

Expected Records Are Missing

The AI assistant can only return Records available to the signed-in user through the exposed View.

Check the user's Softools permissions and the Fields available in the View.

 

Data Lens Results Are Incomplete

Check:

  • Which Data Lens Fields have been configured.
  • Whether the relevant data exists in those Fields.
  • Whether the Fields needed in the response are exposed through the MCP View.

 

The AI Assistant Keeps Asking for Approval

This may be controlled by the AI assistant's own tool-permission settings.

For example, Claude allows individual tools to be configured to run automatically, require approval, or be blocked.

 
 

Summary

MCP provides a controlled, read-only way for supported AI assistants to work with live Softools information.

An Advanced App Builder approves each AI assistant once for the site and decides which Views are exposed. One connection can serve multiple Apps, Views and users, including Views published later. Each user signs in with their own Softools account and grants read access, so their existing permissions continue to determine which information they can access.

Once connected, users can query structured View data, ask questions relevant to their own Softools identity and, where enabled, use Data Lens to perform semantic searches across configured App information.

Comments

0 comments

Please sign in to leave a comment.