Concepts

MCP & tools

An agent with no access to your data is blind. MCP (Model Context Protocol) is the open standard Anthropic created to cleanly plug an agent into your existing tools.

What MCP is

MCP is a protocol, like HTTP or SMTP. It defines how an AI agent and an external tool (Notion, Gmail, your CRM) should talk to each other in a secure and standardized way.

Concretely: Anthropic offers a catalog of MCP servers, both official and third-party. When you enable an MCP server for your account, your agents can use it as an additional tool.

Standard tools (no MCP)

Even before talking about MCP, your agents have access to Anthropic's built-in tools:

  • read, write — files in the agent's environment
  • bash — running sandboxed shell commands
  • web_search — web search
  • web_fetch — fetching the content of a URL

For 60-70% of use cases, these tools are enough. It's when you want to integrate your data or your workflows that MCP becomes indispensable.

Popular MCP servers

  • Notion — read and write in your Notion pages and databases. Ideal for agents that consult internal documentation or create tickets.
  • Gmail — read your emails, draft replies, sort them. Automatic triage, prospect follow-up.
  • GitHub — read repos, comment on PRs, create issues. Code review, doc generation.
  • Google Drive — read and write files, sheets, docs. Reporting, archiving.
  • Slack — read and send messages in channels or DMs. Notifications, conversation summaries.
  • Linear / Jira — ticket management, sprint tracking, retro generation.

The full catalog is in the official Anthropic docs (MCP section).

How to enable an MCP

Enabling an MCP server happens on the Anthropic side, not at Forgent. Steps:

  1. Go to the Anthropic console, MCP / Connectors section.
  2. Choose the service to connect (Notion, Gmail...). You'll be redirected to the OAuth authentication of that service.
  3. Authorize the requested permissions (read-only, read + write, etc.).
  4. Back on the console, the MCP is enabled for your account. All your existing and future agents can use it.
Why on the Anthropic side?Because the OAuth permissions and tokens have to stay in the environment where the agent runs — otherwise you'd have to hand us your Gmail, Notion, etc. access. That would be a huge security risk. By going through Anthropic, your access stays between you and them.

Permissions and security

When an agent uses an MCP, it has exactly the permissions you granted at the time of the OAuth flow. If you gave Notion read-only permission, the agent won't be able to write anything, period.

You can revoke an MCP permission at any time from the Anthropic console or directly at the provider (Google, GitHub, Notion). The agent loses access immediately.

Principle of least privilegeAlways grant the most restrictive permissions possible. An agent that just needs to read one specific Drive folder shouldn't have access to your entire Drive. Anthropic often offers granular permissions: use them.

Custom MCP

If you have an internal API or a proprietary tool, you can write your own MCP server (it's an open protocol). For now, Forgent doesn't offer a UI to register a custom MCP — it's on the roadmap.