Summary
Pi has many built-in tools for working with Planhat data. MCP Plugins add tools from external applications - such as searching Slack or reading a Google Doc - so Pi can bring that context into your Sessions
MCP stands for Model Context Protocol, an open standard for connecting AI assistants to tools. Planhat provides a library of 20+ pre-configured Plugins for popular applications, with Slack, Google Drive and GitHub as featured starting points. You can also connect any other MCP server as a custom MCP connection
An Admin sets each Plugin up once. Depending on the connection type, each User may then need to connect their own account with a couple of clicks (see next point)
Two connection types are available: "Central" (one shared connection for the tenant) and "Individual" (each User connects their own account and sees only what they have access to). Individual is recommended for Slack, Google Drive and any application with personal data
Everything Pi does through a Plugin is logged in Planhat, and Users can control which tools Pi may use on their behalf
Who is this article for?
Anyone who would like to understand what MCP Plugins are in Planhat OS, and what they make possible
Managers and Admins deciding whether and how to roll MCP Plugins out to their team
Series
Planhat OS - MCP Plugins: overview ⬅️ You are here
Article contents
Introduction
Pi, the main AI Agent in Planhat OS, comes with a large set of built-in tools for working with your Planhat data. Much of the context a customer team relies on, however, may live in other systems - messaging software, document storage, ticketing apps, and so on.
MCP Plugins let Pi connect to those external systems and use them as tools, in the flow of a Session. Pi can retrieve information from a connected application, act on it together with your Planhat data, and - where allowed - write back.
Planhat provides a library of pre-configured Plugins for popular applications (20+ at launch, including Slack, Google Drive and GitHub), and additional Plugins continue to be added. If the application you need is not in the library, you can connect its MCP server as a custom MCP connection.
This makes Pi a bridge between Planhat OS and the rest of your tech stack. It is also one half of a broader interoperability story: just as external AI agents can connect to Planhat through Planhat's MCP server, Pi can now call out to other MCP-enabled tools.
Click the image to view it enlarged
📌 Important to note
"Tools" in this context should not be confused with the Tools within Planhat OS (e.g. Data Explorer, Settings, and so on).
What is MCP, and what is an MCP Plugin?
MCP (Model Context Protocol) is an open standard that lets an AI assistant discover and use "tools" offered by another application. Many applications now expose an MCP server - a catalog of things an AI is allowed to do, such as "search messages" or "read a file" - and AI assistants that speak MCP can plug into them.
An MCP Plugin in Planhat is a configured connection between Pi and one such external application. Planhat provides templates for popular applications, so setting one up mostly means supplying credentials from the external application and choosing how Users should authenticate. For applications without a template - including your own internal MCP servers - you can add a custom MCP connection by entering the server details yourself. Once the Plugin is connected, its tools appear to Pi alongside its built-in Planhat tools, and Pi chooses between them based on what you ask.
📌 Important to note
Planhat supports MCP in both directions, and it's useful to point out the difference:
MCP Plugins (described in this article): Pi, inside Planhat, uses tools from external applications
Planhat's MCP server: an external AI such as Claude or ChatGPT uses Planhat's tools to read and act on your Planhat data. See:
What can Pi do with an MCP Plugin?
Exactly what Pi can do with an MCP Plugin depends on the tools the external application provides. The common pattern is: retrieve something from the external application, then do something useful with it in Planhat.
Some typical examples are:
Application | Example | Typical outcome |
Slack | "What is the latest in the Slack channel for Acme?" | A summary of recent customer intel, without leaving Planhat |
Slack | "Summarize the Acme channel and save it as a note on the Company" | Slack context becomes part of the customer record for the whole team |
Google Drive | "Find the files related to Acme in Drive" | Account plans, QBR decks and success plans located in seconds |
Google Drive | "Read the Acme mutual action plan and list the open next steps" | The AE's plan feeds directly into CS work |
Google Drive | "Add today's outcome to the Acme success plan" | Where write tools are enabled, Pi updates the shared document |
GitHub | "Is there an open GitHub issue for the bug Acme reported?" | Engineering status is visible to CS without switching tools |
Pi combines these with everything it already knows from Planhat, so a single request can draw on Slack, Drive and your Planhat data at once. If a routine becomes standard - say, a pre-call brief that always pulls from the same places - it can be saved as a Skill so every CSM runs it the same way.
How it works
Roles: Admin and User
Note that for "Admin" here we are referring to a user of Planhat who's an Admin, rather than a Planhat staff member (a Super Admin).
| Admin | User |
Does what? | Creates the MCP Plugin from a template, registers an OAuth app in the external application, adds the credentials, chooses the connection type, and manages the Plugin over time | Connects their own account with a couple of clicks (for Individual connections), optionally chooses which tools Pi may use, and then uses the MCP Plugin simply by asking Pi in Sessions |
Where are MCP Plugins set up / used? | MCP Plugins Tool, plus the external application's admin console | MCP Plugins Tool, then any Session |
How long does setup take? | Typically 15-30 minutes per application, plus any internal approval | About 30 seconds, once per application (for Individual connections) |
Detailed guide article |
Connection types
When creating a Plugin, the Admin chooses one of two connection types:
Individual - each User connects their own account. Pi sees only what that User can see. This is the recommended (and, for most teams, the only sensible) choice for Slack, Google Drive and anything else containing personal data such as direct messages or private files
Central - one shared connection for the whole tenant. Every User works through the same account in the external application, and that application cannot tell Users apart (although Planhat's own logs can). Suitable for shared, non-personal sources such as a company knowledge base
Click the image to view it enlarged
Security and control
OAuth first, with alternatives where needed. MCP Plugins support three authentication methods: OAuth, API token, and username/password. Most pre-configured Plugins use OAuth, which is the recommended method under the MCP specification, and offers the strongest security: Users authorize on the external application's own consent screen, Planhat never handles their password, and access can be revoked at any time
Your access, nothing more. With Individual connections, Pi cannot reach anything the User could not open themselves
Tool-level control. Users can switch individual tools on or off - for example, allow reading but not writing
Audit trail. Planhat records every tool call, including who made it and when
How to get started - process summary
Decide on your first use case - for most organizations, this is likely to be Slack (e.g. for customer channels) or Google Drive (e.g. for account plans)
Line up access in the external application (for example, you'll need someone who can create a Slack app or a Google Cloud OAuth client, for those applications), and start any internal security review (if required) early
Follow the MCP Plugins Admin guide to create the MCP Plugin, and test it with your own account
Share the MCP Plugins User guide with your team
Optionally, train Pi by creating a Skill, so the procedure is consistent across the team