Delv
IDEby Continue Dev★ 4.3

Continue

Open-source VS Code / JetBrains extension for custom AI assistants. Supports MCP servers, local models, and agent-mode workflows.

Homepage MCP docs Source
B
Safety & Trust

Delv Safety Grade: B

Score 74/100 · assessed 2026-04-19

Maintainer65
Permissions55
Supply chain85
Transparency88
Incidents100

Continue is an open-source IDE extension from a small but active startup team. The project has strong transparency with excellent documentation and active development, distributed through official VS Code and JetBrains marketplaces with reasonable supply chain practices. However, as an MCP client host, it inherits the full permission scope of any MCP server you connect to it. The extension itself requires filesystem access to your workspace and can execute code through connected MCP servers. The maintainer is a small commercial entity (Continue Dev) rather than a major vendor, which affects the bus factor. No security incidents are known. The MCP implementation is partial (tools and resources only, no prompts or sampling yet), which actually limits attack surface. Configuration through YAML or UI is more user-friendly than raw JSON but still requires understanding what permissions each MCP server requests.

Green flags

  • Open source with active development and 13k+ GitHub stars
  • Distributed via official VS Code and JetBrains marketplaces
  • Excellent documentation including MCP-specific guides
  • UI-based configuration reduces manual YAML editing errors
  • No known security incidents or CVEs

Red flags

  • Inherits full permissions of any connected MCP server without sandboxing
  • Small startup maintainer with limited bus factor compared to major vendors
  • Partial MCP implementation may lead to compatibility issues with some servers
  • Agent-mode workflows can execute arbitrary code through MCP tool calls

Permissions requested

Read filesWrite filesShell executeOutbound networkExternal LLM callRead env
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

MCP capabilities

  • Tools
  • Resources
  • Prompts
  • Sampling

Platforms

macOSWindowsLinux

Config location

~/.continue/config.yaml or the Continue UI

Review

Continue is the most IDE-native way to bring MCP servers into your editor. It's an open-source extension for VS Code and JetBrains that turns your IDE into a full-featured AI assistant host, with MCP support baked in alongside local model support and agent-mode workflows. The MCP implementation is solid. Tools and resources work as expected, though prompts and sampling aren't supported yet. Configuration lives in ~/.continue/config.yaml or through the UI, which is more approachable than editing JSON by hand. I've been running it with filesystem and database MCP servers for a few weeks, and the integration feels tighter than Claude Desktop because everything stays in the editor. No context-switching to a separate chat window. What sets Continue apart is flexibility. You can swap between OpenAI, Anthropic, local models via Ollama, or even custom endpoints. The agent mode lets the assistant execute multi-step tasks with tool calls, which pairs well with MCP servers that expose database queries or API actions. I've used it to scaffold components while querying a Postgres MCP server for schema details, all without leaving the IDE. The UI is clean but opinionated. The chat panel sits in the sidebar, which works for quick queries but feels cramped for longer conversations. The inline edit mode is better for refactoring, but it took me a few days to learn when to use which. Documentation is decent, though the MCP-specific setup guide assumes you already understand MCP server architecture. Performance is fine with a couple of servers, but I've noticed sluggishness when running three or more simultaneously. The extension doesn't always surface MCP server errors clearly, so you'll need to check logs if something breaks. Configuration reloads require a full IDE restart, which is annoying during setup. Continue shines if you want MCP servers embedded in your daily coding workflow and you're comfortable tweaking YAML. It's less polished than Claude Desktop for pure chat, but far more powerful for developers who want their AI assistant to live where they work. I'd reach for this over standalone clients when I need tight integration with my codebase and don't mind trading some UX polish for control.
Verdict

Best MCP client for developers who want AI assistance embedded in their IDE with full control over models and servers. Skip it if you prefer a polished chat-first experience or don't want to manage YAML config files.

Good at

  • Native IDE integration means no context-switching to external chat apps
  • Supports local models, custom endpoints, and multiple providers alongside MCP
  • Agent mode enables multi-step workflows with tool calls
  • Open-source with active development and community
  • Config via UI or YAML gives flexibility for power users

Watch out

  • No support for MCP prompts or sampling capabilities yet
  • Chat panel feels cramped for longer conversations
  • Performance degrades with multiple MCP servers running
  • Config changes require full IDE restart
  • Error messages from MCP servers aren't always surfaced clearly