home / ai-agents / continue

Continue

Official · Continue
32k

Coding-agent CLI for repo-aware chat, headless tasks, diff-based reviews, and Continue-managed assistant or MCP configs.

$curl -fsSL https://raw.githubusercontent.com/continuedev/continue/main/extensions/cli/scripts/install.sh | bash
Language
TypeScript
Stars
31,714
Category
AI & LLM Tools
Agent
Ready
Agent Compatibility
JSON Output
Agent Skill
MCP Support
AI Analysis

Continue CLI (cn) is a terminal coding agent that can run interactive sessions or one-shot headless tasks against a local checkout. It shares assistants, rules, and MCP configuration with Continue's hosted platform and editor extensions.

What It Enables
  • Run repo-aware coding tasks that read files, execute commands, edit code, and resume prior sessions from the terminal.
  • Review local diffs with cn review, get text or JSON reports, emit patches, and optionally apply suggested fixes to the working tree.
  • Use Continue-managed assistants, rules, and MCP servers from the shell, plus inspect PR check results with cn checks or expose a remote agent session with cn serve and cn remote.
Agent Fit
  • Headless -p, TTY-less safeguards, cn ls --json, and cn review --format json make it usable in scripts, CI jobs, and editor-driven loops.
  • A lot of the value still flows through an LLM conversation loop, so results are less deterministic than narrow service CLIs and often need prompt, model, or permission tuning.
  • Fits best as a higher-level repo automation layer when a skill pins the config, allowed tools, review agents, and any optional MCP servers.
Caveats
  • Unattended use needs authentication up front through cn login, CONTINUE_API_KEY, or an Anthropic key, and default mode is still interactive.
  • Headless --format json guarantees parseable JSON, but chat responses may be wrapped model text rather than a fixed schema unless you control the prompt and workflow.