Atlas supports the Model Context Protocol (MCP): it connects to MCP servers and exposes their tools to the agent, so you can bring your own tools and data sources into the terminal. Combined with its plugin system, MCP makes Atlas extensible without leaving the shell in 2026.
What MCP adds
The Model Context Protocol is an open standard for connecting AI agents to external tools and data. Atlas connects to MCP servers and converts their tools into agent tools, so the model can call them like any built-in.
That means you can wire Atlas to issue trackers, databases, documentation, or internal services through MCP, and the agent uses them as part of a task. Because the tools are imported into the same registry as built-ins, they participate in planning and permissions like everything else.
Safe by default
MCP servers can read and write outside your repository, so Atlas keeps them behind the same permission model as every other tool. Each MCP tool call is gated against allow, ask, and deny rules.
This matters: extending an agent should not weaken its guardrails. With Atlas, an MCP tool that writes data still prompts for approval unless you explicitly allow it. You get extensibility without giving up the auditability that makes the agent safe to run.
Plugins and MCP together
Alongside MCP, Atlas has a plugin system that contributes tools and hooks into agent lifecycle events. Together they cover both first-party extensions and standard third-party servers.
Use plugins for tools you build for Atlas directly, and MCP for the growing ecosystem of standard servers. Either way, you shape Atlas around your stack, and you can read the open-source code to see exactly how a tool is wired in.
Frequently asked questions
- Does Atlas support Model Context Protocol?
- Yes. Atlas connects to MCP servers and exposes their tools to the agent so you can extend it with your own tools and data sources.
- How does Atlas import MCP tools?
- Atlas converts each MCP server tool into an agent tool and registers it alongside built-in tools, so it participates in planning and permissions.
- Are MCP tools permission-gated?
- Yes. Every MCP tool call runs behind Atlas's allow, ask, and deny permission model, just like built-in tools.
- What is the difference between plugins and MCP in Atlas?
- Plugins are first-party extensions that contribute tools and hooks; MCP connects Atlas to standard third-party servers. Atlas supports both.
Try Atlas in your terminal
The terminal-native AI coding agent. Open source, single binary.
Install AtlasRelated guides
Atlas vs Cursor: terminal AI coding agents compared (2026)
A grounded 2026 comparison of Atlas and Cursor across workflow, change review, extensibility, and pricing for developers choosing an AI coding agent.
Atlas vs GitHub Copilot: agent vs autocomplete (2026)
How Atlas, a terminal-native AI coding agent, compares to GitHub Copilot's editor-embedded autocomplete and chat in 2026.
Atlas vs Aider: terminal AI coding tools compared (2026)
Atlas and Aider are both open-source terminal AI coding tools. A 2026 comparison of their interfaces, review models, and extensibility.
The best alternative to Cursor for terminal users (2026)
Looking for a Cursor alternative that lives in your terminal? Atlas is an open-source AI coding agent that plans, diffs, and gates every change.