Use your SecurityTrackr account with MFA set up. MCP is available on every SecurityTrackr plan, including Free, within your organisation's limits. Your agent provider's access rules and charges apply separately.
Remote MCP server URL:
https://app.securitytrackr.com/api/mcpConnect a Copilot Studio agent
This walkthrough uses Copilot Studio's MCP onboarding wizard, then publishes the agent to Microsoft 365 Copilot. You need access to a suitable Copilot Studio environment and permission to add tools and publish agents.
- Open your agent's Tools page. Choose Add a tool → New tool → Model Context Protocol.
- Name the server SecurityTrackr, enter the URL above, and describe it as a security register for finding and managing security observations.
- Select OAuth 2.0, then Dynamic discovery. This Copilot Studio flow uses DCR, which SecurityTrackr offers for clients that do not support CIMD. Its registration endpoint is advertised through OAuth metadata.
- Create the server, continue, and create a connection. Complete SecurityTrackr sign-in and consent, then add the tool to the agent.
- Try the read-only check in the agent's test experience before publishing.
If your environment blocks the connector, check its Power Platform data policies with your administrator.
Source: Microsoft's MCP onboarding guide.
Before sharing: whose access does the agent use?
SecurityTrackr authorises each MCP call as the account that approved the OAuth grant, limited by that account's current role and the permissions granted to the connection. It does not automatically use the identity of the person chatting with the published agent.
If you share an agent that uses one shared connection, people who can use that agent may be able to read register data or request changes through the grant owner's access. SecurityTrackr records those actions under the grant owner and the connected application. Publishing the agent does not create a separate SecurityTrackr identity for each person using it.
Start with read-only access for a reporting agent. For an agent that contributes findings, add only the creation and update permissions its work needs. Leave risk acceptance and observation deletion off. The consent screen leaves these high-impact permissions unselected by default, but preselects other permissions requested by the client: review and untick anything unnecessary.
Before broad sharing, check which connection the published agent uses, who can invoke it, and what that connection allows. Test with another intended user and a low-privilege grant. Do not assume a colleague's own SecurityTrackr role limits a shared connection granted by someone else.
Consider a dedicated account for the agent
An agent can use a dedicated SecurityTrackr account with its own email address. Invite that address into your existing organisation, rather than creating a separate organisation. This is a normal user account and counts towards your plan's member limit.
Choose Auditor for read-only work, or Contributor when the agent needs to create and update observations. These roles also limit what can be granted at consent; you can narrow the connection's permissions further. A Contributor can read the register, so this does not restrict it to selected observations.
Have a responsible person complete the invitation, email verification, MFA setup and OAuth consent for the dedicated account. Keep its sign-in credentials and recovery methods under your organisation's control. The agent uses the resulting OAuth connection; it does not need the password or MFA secret in its instructions. Plan for reauthorisation when the connection expires, and revoke it under Account → Connected agents when it is no longer needed.
Make the agent available in Microsoft 365 Copilot
Publish the agent, then open Channels and configure the Microsoft 365 and Teams channel. Enable availability in Microsoft Copilot and add the channel. Install it for yourself first, then share it with the intended users or request organisation-wide approval.
Approve access in SecurityTrackr
Complete sign-in in the browser and review the requested permissions. For a first read-only check, keep observations.read and remove permissions you do not need. Choose one-time access for 15 minutes, or a persistent connection when offered. Longer access and sensitive permissions require fresh MFA verification.
MCP uses OAuth. Prefer Client ID Metadata Documents (CIMD) when your client supports them. SecurityTrackr retains Dynamic Client Registration (DCR) for compatibility with clients that do not. Both require your consent before accessing the register. A SecurityTrackr REST API token is not a substitute.
Check the connection
Enable the connection in your agent and ask for an overview of your register:
Use SecurityTrackr to summarise the top 5 unhandled risks in our register. Include their observation identifiers and explain their priority. Do not change any records.
Check that the agent actually called SecurityTrackr and that its answer matches records you can open. An empty result can be valid. Confirm the connection under Account → Connected agents, where you can also disconnect it.
If the connection fails
Check the server URL, finish the browser sign-in and MFA prompts, and confirm that the connection is enabled in the conversation. A permission error can mean your role or approved scopes do not allow that action. Reconnect when access expires; refresh the client's tool list if it still shows old tools.
This setup is based on the linked provider documentation. It is not an end-to-end certification of every client version or workspace configuration.
Read the shared MCP guide for permissions, duplicate review, background AI jobs and revocation.