On this page
What is an MCP server?
The Model Context Protocol (MCP) is an open standard for connecting AI assistants to outside tools and data, described at modelcontextprotocol.io. An MCP server publishes a set of tools, such as "list tickets" or "create a Planner item", that a compatible AI client can call on your behalf. A remote MCP server runs on the vendor's side, so you add its address to your AI client instead of installing anything.
For an MSP, this means asking an AI assistant a question like "which clients have device warranties expiring in the next 90 days?" and having it answer from your portal data rather than from a pasted spreadsheet.
What the MSPortal MCP server can do
The catalog groups 359 operations into 114 public tools, which keeps the full set under the 128-tool limit some clients have. Tools cover tickets, devices, reporting, Planner, goals, budgets, meetings, compliance, projects and project documents, financial data, cloud services, security, backup, surveys, training, onboarding and more. Read tools are named like query_planner, and change tools like manage_planner.
Some examples from the MCP tools reference:
- list and read tickets, and add a note or change a status, written to ConnectWise, HaloPSA, Autotask or Syncro through the PSA's API;
- create a ticket through the active PSA integration;
- read devices, backup and compliance data for a client;
- draft compliance notes;
- create Planner items, meetings and reports.
What a session can actually do depends on the server version, the areas you approved, your role, your company access and your integrations. A client user's connection never gets MSP-only tools.
How MCP sign-in and permissions work
- No API keys. Clients connect over Streamable HTTP and sign in with OAuth 2.1 and PKCE in the browser.
- Your own access. Every tool call is checked against your role, tenant and company scope. Broader OAuth scopes cannot widen it.
- Choose areas at consent. On the sign-in page you tick only the areas the client should use.
- Permission to connect. Users need the MCP Server write permission, which the MSP administrator grants per role.
- Revoke anytime. Settings > MCP Server lists your connected clients with a Revoke button. An access token already issued can work for up to about 15 minutes afterwards.
Data that a tool returns goes to the AI client you connected and is handled under that service's terms, so connect only the areas you need.
Confirmation before changes
Most changes use a two-step flow. The AI client first asks for a change and MSPortal returns a proposal showing the company, records and effects, and nothing is saved. The change happens only when the client calls again with the confirmation ID, within ten minutes. A few actions run at once, such as marking a notification read, and generating compliance note drafts does not save them. Check the record afterwards, since a multi-step workflow is not rolled back as a unit.
Which AI clients work with MSPortal
The MCP setup guide has instructions for Claude Code, Claude Desktop and web, Cursor, GitHub Copilot in VS Code, OpenAI Codex, Gemini CLI and Windsurf, with one-click install for Cursor and VS Code. The pricing page also lists ChatGPT. Any client that supports remote Streamable HTTP MCP servers with browser OAuth sign-in should work. An optional skill file, Ron, guides an assistant through client briefings, risk reviews and QBR preparation.
The MSPortal Public API
The REST Public API v1 is for integrations and scripts:
- Authentication: an API key sent as a Bearer token. Keys are created in Settings > Integrations > API Access, carry scopes and inherit the creating user's permissions.
- Coverage: about 80 endpoints across companies and company users, users, tickets, devices, compliance, budgets, calendar meetings, goals, Planner, reports and report schedules, dashboards, surveys, training, automation rules and analytics.
- Limits: requests are rate limited per key, with limit headers on each response, and every response has a request ID for support.
- Documentation: an OpenAPI 3.0 specification, published as the API Reference tab on docs.msportal.ai.
Tickets are read-only in the Public API: you can list and read them. Creating tickets and adding notes happen in the portal or through the MCP server.
MCP server or API: which should you use?
Use the MCP server when a person wants to work with portal data through an AI assistant, such as preparing for a client meeting or asking questions across clients. Use the Public API for scheduled scripts and system-to-system integrations that need predictable endpoints and no human in the loop. The AI page covers the built-in assistant, and what AI does in MSPortal explains where it helps.
Related questions
Does the MSPortal MCP server need an API key?
No. AI clients connect to the MSPortal MCP server with browser-based OAuth 2.1 and PKCE sign-in. There are no keys or secrets to copy, store or rotate.
Can an AI assistant change data in MSPortal without approval?
Most changes cannot. MSPortal's MCP server returns a proposal first, and the change is saved only when the client confirms it within ten minutes. A few small actions, such as marking a notification read, run at once.
Can client users connect an AI assistant to MSPortal?
Only if the MSP gives their role the MCP Server permission. A client user's connection is limited to their own company's data and never receives MSP-only tools.
Is API access included in every MSPortal plan?
Yes. The pricing page lists REST API access and the MCP server as included in every plan. See how much MSPortal costs.
Can MCP manage MSPortal projects?
Partly. MCP tools read projects and can write updates, milestones, deliverables, action items, status, owners and priorities after confirmation. See project management for what stays in the app.