Skip to main content

PEAK MCP Server

Connect PEAK to Claude, ChatGPT, or Microsoft Copilot using the PEAK MCP Connector. Ask questions about urgent alerts, open actions, and equipment issues across your portfolio in plain English.

Written by Howie

Getting started with the PEAK MCP Connector

Bring PEAK into the AI assistant you already use. The PEAK MCP Connector securely extends your portfolio into Claude, ChatGPT, Microsoft Copilot, and any other AI client that supports the Model Context Protocol (MCP).

Engineering and operations managers can ask questions in plain English and get structured answers drawn directly from live PEAK data. Triage urgent alerts, chase open actions, and check site health from inside the tool you're already working in. No dashboard hopping. No context switching. No exporting CSVs to ask follow-up questions.

MCP is an open standard backed by all major AI clients, and the PEAK MCP Connector works with any of them using the same connection URL. This article walks through two setup flows tested end to end:

  • Anthropic Claude; and

  • Microsoft 365 Copilot and Teams

Other clients such as OpenAI ChatGPT use the same URL but their setup steps differ.

Claude PEAK MCP Connector screenshot

Who can use it

You will need:

  • An active PEAK account with access to one or more sites

Permissions are inherited from your existing PEAK login. You will only be able to see and query sites your PEAK user already has access to.

Each AI client then has its own requirement:

  • Claude: a Claude account (Free, Pro, Team, or Enterprise). You can test it yourself with a free account in a couple of minutes. See Setup in Claude.

  • Microsoft 365 Copilot and Teams: a Microsoft 365 Copilot Studio license to build the agent and a Microsoft 365 Copilot license to use it. The build is typically done once by an IT admin, then shared to the rest of the team. See Setup in Microsoft 365.

MCP Connection URL

Environment

Server URL

Production

The same URL is used by every AI MCP client.

Setup in Claude

  1. Navigate to claude.ai or open the Claude desktop app.

  2. Go to Settings > Connectors > Add custom connector.

  3. Configure the integration:

    • Name: PEAK

    • URL: https://api.cimenviro.com/mcp

  4. Complete the PEAK OAuth authorization when prompted.

  5. Start asking questions about your portfolio.

Claude connector permissions screenshot

Setup in Microsoft 365 Copilot and Teams

Microsoft 365 requires reusable agents to be built in Copilot Studio. Copilot Studio connects to the PEAK MCP server through Tools, then shares the agent through Microsoft 365 Copilot and Teams so your whole team can use it.

Asking your PEAK Agent operational questions inside Microsoft 365 Copilot

Microsoft 365 prerequisites:

  • A Microsoft Copilot Studio license for the person building and publishing the agent.

  • A Microsoft 365 Copilot license for every person who will use the agent. Microsoft requires a separate Copilot license to use third-party agents.

  • Your PEAK login, used once to authorize the connection to PEAK MCP.

Part 1: Build the agent in Copilot Studio

  1. Click Agents > Create blank agent and give it a name, for example PEAK Agent.

  2. Select your agent's model, such as GPT or Claude model. Pick a reasoning-capable model for the best results.

Create new agent in Copilot Studio

Select your agent's AI LLM model

Part 2: Connect the PEAK MCP server

  1. Open the Tools tab and click Add a tool.

  2. Select Add new MCP and complete the required fields:

    • Server name: PEAK MCP

    • Server description: Connect Microsoft 365 agents to CIM PEAK MCP data

    • Server URL: https://api.cimenviro.com/mcp

    • Authentication: OAuth 2.0

    • Type: Dynamic discovery

  3. Click Connection > Create new connection > Create. You will be asked to log in to your PEAK account.

  4. Once authenticated, click Add and configure. All PEAK MCP tools will now be available to the agent.

Navigate to Tools > Add Tool > Add MCP server

Create new connection and authenticate with your PEAK user login credentials

Part 3: Publish and share to Microsoft 365 Copilot and Teams

  1. Click Publish to publish your agent.

  2. Go to the Channels tab, select Microsoft 365 and Microsoft Teams, and click Add channel. Turn the channel on.

  3. In the Microsoft 365 and Microsoft Teams panel, check Make agent available in Microsoft 365 Copilot.

  4. In Agent preview, click Edit details. Fill in a short and long description, and under Teams settings check all Teams sharing permissions

  5. Click Save.

Publish and go to Channels to share

In Channels select Microsoft 365 and Teams and click Add channel

In Agent preview click Edit details and enable sharing in Teams

Part 4: Add and use the agent

You can reach the published agent from either Microsoft 365 Copilot or directly in Teams.


From Microsoft 365 Copilot:

  1. Open a New chat.

  2. Type @PEAK in the chat, find your PEAK Agent and select it to connect

In Microsoft 365 Copilot, New Chat > @PEAK to find agent and select to add

From Microsoft Teams:

  1. In Copilot Studio Agent preview, click See agent in Teams, then click Add to add the agent to Microsoft Teams.

In Copilot Studio Agent preview click See agent in Teams, then Add to Teams

Part 5: First-time PEAK authentication

The first time anyone asks a question, the agent needs to verify their PEAK credentials.

  1. The agent will reply with a prompt to connect. Click Open connection manager and log in to PEAK.

  2. Return to the chat and click Retry.

After this one-time step, questions return detailed answers directly in the chat.

First time asking a question, open connection manager to authenticate

Detailed conversation reply and AI agent assistance

Sharing the agent with teammates

To give others access, click Share app and message a teammate so they can add it to their own Teams account. Each teammate may need their own Microsoft 365 Copilot license to use third-party agents, and will go through the same one-time PEAK authentication.

Share agent to users or groups

What you can do

  • Find your portfolio. Look up clients, sites, equipment, and equipment types across your account. This is the starting point for any question that needs to be scoped to a specific building or piece of plant.

  • Track issues. Search alert tickets, which are triaged fault detections from the PEAK rules engine, and action tickets, which are escalated tickets assigned for follow-up.

  • Update actions. Comment on, reassign, and update the status of open actions. Use your AI client to help draft responses and close out actions.

  • Update rule thresholds. Users with engineering permissions can bulk-tune rule settings, including priorities, thresholds, schedules, start/stop rules, and more.

  • Query live trend history. Analyze equipment and sensor trend data over time to drill deeper into alerts or develop business cases. Note: interval data queries typically analyze more data and may take longer and use more tokens.

Example queries

Try these in your Claude chat once the connector is set up.

What issues need attention at my sites?
Find all chiller issues at site: [enter site name]
What actions need following up across my sites?

The assistant will use the PEAK tools to look up your sites, run the relevant ticket searches, and return a structured summary you can act on. From there you can drill in further: ask about a specific assignee, narrow to a single equipment type, or pull the full detail on any alert or action.

Extending with skills

Skills are reusable prompt packs that turn a single PEAK MCP query into a structured workflow. Rather than writing the same prompt every time, you install a skill once and invoke it with a slash command.

Featured skill: peak-ops

The peak-ops skill is a portfolio operations dashboard for the PEAK MCP. It produces structured rollups of priority alert tickets, assignee load, and unresolved issues across your sites, and supports follow-up drilldowns by site, equipment type, or equipment name within the same session.

Skill repo: https://github.com/cim-data-engineering/skill-peak-ops


Featured skill: peak-close-audit

The peak-close-audit skill is a sensor-verified QA of PEAK alerts and actions. Give it a list of alerts and actions and it assesses which actions can be confidently closed. It returns a summary table with a verdict (✅ Close / 🚨 Keep / ⚠️ Investigate), then offers to batch-close based on your confirmation.

Skill repo: https://github.com/cim-data-engineering/skill-peak-close-audit

Install steps (the same for every skill):

  1. Visit skill repo link in your browser (see above)

  2. Click the green Code button and select Download ZIP.

  3. In Claude, go to Customize > Skills > Upload skill.

  4. Drag the downloaded zip into the upload area.

  5. In any chat, invoke the skill with its slash command (/peak-ops or /peak-close-audit).

Once invoked, the skill stays active for the rest of the session, so you can keep asking follow-up questions without re-running the slash command.

Troubleshooting

Token expiry.

If the connector starts failing or returning authentication errors, your PEAK token has likely expired.

  • In Claude: Go to Settings > Connectors, disconnect and re-connect the PEAK connector, then re-authenticate.

  • In Copilot or Teams: when the agent prompts you, click Open connection manager, log in to PEAK again, then click Retry.

Missing sites or data.

You can only see sites your PEAK login already grants access to. If a site is missing, check your PEAK permissions before raising a support ticket.


A teammate cannot use the shared agent

Third-party agents may require a Microsoft 365 Copilot license depending on your Microsoft enterprise licensing agreement. If a teammate cannot add or use the PEAK Agent, confirm their Copilot license with your Microsoft IT Admin before troubleshooting further.

Security and data access

The PEAK MCP uses OAuth 2.0 for authentication and inherits the permissions of your existing PEAK account. No additional data access is granted through the connector. The AI client (Claude, ChatGPT, or other) processes your queries and responses according to its own privacy and data retention policies, so review your organization's policies for AI tool usage before connecting.

Did this answer your question?