MCP

Your codebase is only part of the picture. Real work also lives in issue trackers, logs, databases, and docs. MCP is how X4RGE connects the agent to those systems so it can use live data during a task — not just what you paste into chat.

Think of MCP as plugging the agent into your stack: you connect a service once; the agent can reach it when a conversation needs it. Most connections come from the plugin marketplace. Your organization decides which integrations are available.

What is MCP

MCP (Model Context Protocol) is an open standard for linking AI agents to external tools. In X4RGE, each connection is an MCP server — a bridge to one service (for example Linear or Datadog) that exposes actions the agent can take: search issues, pull error reports, read schema metadata, and similar tasks.

You do not use MCP directly. You install or enable a connection, then describe what you want in chat. The agent decides when to use that connection, and sensitive actions follow the same approval flow as editing files or running commands.

In practice

When you ask…	What MCP enables
"Why is checkout failing in prod?"	The agent can query Sentry or Datadog instead of guessing from repo code alone
"File a bug for this with repro steps"	The agent can create or update a ticket in Linear or Jira
"What columns does our users table have?"	The agent can inspect a connected database or Supabase project
"Summarize what we decided in Notion"	The agent can pull from connected docs when your team has that plugin

If a service is not connected, the agent still helps using your repo and what you provide in the message — MCP adds reach into the tools you already rely on.

Why use MCP

Without a connection	With MCP connected
Paste error logs or ticket text into chat	The agent pulls recent errors or issues from the source
Describe what happened in production from memory	The agent queries monitoring tools with your approval
Manually file follow-up work after the agent finishes	The agent can create or update tickets when you ask

Connections keep the agent aligned with what is true right now, not a snapshot you copied earlier in the thread.

What you can connect

What you can enable depends on your marketplace catalog and org policy. Common examples:

Area	What the agent can help with
Issue tracking	Search, create, or update work in Linear, Jira, and similar tools
Monitoring & logs	Investigate incidents with Datadog, Sentry, Grafana, or Elastic
Databases & data	Inspect schemas or run read-oriented queries on Supabase, Postgres, and others
Docs & collaboration	Pull context from Notion, Confluence, or shared documentation
Cloud & dev platforms	Look up deployments, resources, or project metadata from connected providers

Open Marketplace in the app to see integrations your organization offers.

Get started

Open Marketplace or Tools & MCP
In the desktop app, use Marketplace to install a plugin, or Tools & MCP to view and manage services already connected to your account.
Add a connection
Choose an integration from the marketplace, or ask your admin to enable one for your team.
Sign in
Complete OAuth or enter credentials when prompted. Credentials are stored for the connection — do not paste API keys or tokens into chat.
Confirm it is active
In Tools & MCP, verify the service shows as connected before you depend on it in a task.

Administrators can limit which MCP servers teams may use and how cloud agents reach the network. See Agent security for org-wide controls.

Use in chat

After connections are set up, the agent uses them when they fit the task. You do not need to say "MCP" or pick a server yourself. Mention the service when it matters:

"Check Sentry for checkout errors in the last hour"
"Create a Linear issue for this bug with the repro steps above"
"Query our staging database for users created today"

When the agent calls an external service, you will see that step in the conversation. Actions that change data or reach production systems pause for your approval — review the request before you allow it, the same as for terminal commands or sensitive file edits.

Name the service when you have many connections

If several integrations are enabled, saying Linear, Datadog, or Notion helps the agent choose the right one on the first try.

Stay in control

Approvals — High-impact or write actions wait for your OK in chat. See Tools.
Scoped access — Prefer read-only or limited credentials for investigation work.
Org policy — Admins control marketplace access, models, and cloud network rules.
Turn off what you do not need — Disconnect unused services in Tools & MCP.

Built-in safety, not a bypass

Connected services go through the same controlled path as other agent capabilities. There is no hidden route around approvals for external tools.

Plugins & marketplace

Browse and install integrations that add MCP connections and skills.

Tools

How the agent works in your repo and when each action needs approval.

Agent security

Organization controls for connected services and sensitive operations.