M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
K

Kilo Code

Official
Not a standard server
The open source, all-in-one agentic engineering platform

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #412 of 438
Stars
★ 27.9k Very popular
Runtime
Node.js · Prebuilt binary
Credentials
Optional API key
License
MIT
Last commit
30FMRS · D

Kilo Code is an actively maintained, MIT-licensed open source project from Kilo-Org offering a mature multi-surface (VS Code, JetBrains, CLI, cloud) AI coding agent with multi-model switching, code generation, autocomplete, and CI/CD automation, plus an MCP marketplace for connecting external MCP servers. However, the source material provides no evidence that this repository itself implements or exposes a standard MCP server interface (tool list, transport, connectable MCP clients), so those fields are left empty and it should not be treated as a directly connectable MCP server.

Strongest · Maintenance 10/20 Weakest · Reliability 2/20

Reliability
2/20
Security and permissions
6/20
Maintenance
10/20
Documentation
8/20
Setup experience
4/20
Why each score
Reliability 2/20
No server manifest was provided ([not cached]), and the README describes a general-purpose AI coding agent platform (VS Code/JetBrains extensions, CLI, cloud agent) rather than an MCP server — there is no tool list, no MCP init/handshake evidence, and no description of declared-tool-to-behavior mapping. The README even states the project ships an 'MCP marketplace' to connect to external MCP servers, indicating it functions as an MCP client/host, not the MCP server under review. With no tool definitions or reproducible startup evidence, and per the static calibration rule capping reliability at 12 absent CI/test evidence, this must score very low.
Security and permissions 6/20
No red-line issue is evident (no default unconfirmed destructive/pay/deploy actions, no leaked credentials in examples). However, the README explicitly notes that `kilo run --auto` 'disables all permission prompts and lets the agent execute any action without confirmation,' mitigated only by a text warning to use it in trusted environments — this is a meaningful gap in least-privilege and confirmation disclosure. Because the object is not actually shown to be an MCP server, there is no visibility into any tool-exposure permission model, data boundaries, or credential handling, so the security picture is highly incomplete and must be scored conservatively.
Maintenance 10/20
The repository is MIT-licensed, not archived, and includes CONTRIBUTING.md, RELEASING.md (with separate release processes for VS Code, CLI, and JetBrains), and a CODE_OF_CONDUCT.md, indicating some governance structure. Star and issue counts are discovery signals only and are not used to raise the score. Without concrete evidence of commit cadence, recent release dates, or a security-response channel in the supplied material, sustained maintenance cannot be verified, warranting a mid-range score.
Documentation 8/20
The README documents general product installation (VS Code/JetBrains/CLI across multiple package managers), agent modes, and CI/CD automation reasonably well, with a link to external docs. However, there is no documentation whatsoever addressing an MCP server role: no tool parameter documentation, no authentication flow, no rate-limit/cost disclosure, and no MCP-protocol-specific troubleshooting. Since the review object is nominally an MCP server, this core documentation gap caps the score at a middling level based only on general product docs.
Setup experience 4/20
Install instructions are provided for multiple channels (npm, curl, pnpm, bun, Homebrew, AUR, GitHub release binaries) for installing the Kilo Code agent itself, but none of these describe configuring it as an MCP server inside a mainstream MCP client (e.g., no command/args/env config block for Claude Desktop or similar). There is no evidence the project can be discovered or invoked as an MCP server by a client. Per the static calibration rule, setup cannot exceed 15 without real CI and committed tests covering the server's key paths, and given the absence of any MCP-specific connection steps, a low score is warranted.

Static review · not runListed 2026-08-06

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

  • Developers who want one agent with access to 500+ switchable AI models across editor and terminal
  • Teams setting up automated pull request code reviews
  • Teams wanting an autonomous coding agent in CI/CD pipelines to fix failing tests or complete defined tasks
  • Users who prefer open source tooling billed at provider rates with no subscription markup

Not for

  • Users looking for a standard MCP server that exposes specific tools/resources to connect to (the source material does not show Kilo Code offering this)
  • Fully offline scenarios with no external model provider account or network access
  • Environments where automatic terminal command execution and browser control cannot be trusted or permitted

Required permissions

  • Terminal command execution, used to run and automate tasks
  • Browser control, used to automate web actions
  • Read/write access to local repository files for code generation and edits
  • With `--auto`, the agent can perform any action without confirmation prompts

Risks and side effects

  • `--auto` disables all permission prompts, letting the agent execute any action unconfirmed — the README explicitly warns to only use it in trusted environments
  • Granting terminal and browser control carries risk of unintended or incorrect actions being executed
  • The agent's "self-checking" reviews its own work but does not guarantee correctness, so human review is still warranted
  • The source material does not describe data handling/privacy practices for the third-party model providers it connects to; check official docs before relying on it for sensitive code

Setup

Before you start

Runtime：Node.js · Prebuilt binary

Choose an installation path based on where you work: install the "Kilo Code" extension from the VS Code Marketplace; install the CLI via npm (npm install -g @kilocode/cli), the curl install script, pnpm, bun, Homebrew (Kilo-Org/tap/kilo), or the Arch AUR (kilo-bin), then run kilo in a project directory; install the JetBrains plugin by searching "Kilo Code" under Settings → Plugins; or download a prebuilt binary from the GitHub Releases page for your platform (note the x64-baseline build for older CPUs without AVX, the musl build for glibc-less containers like Alpine, and that the .vsix file is the VS Code extension package, not the CLI). These steps install Kilo Code itself as a coding-agent client; the source material does not include a configuration example for adding Kilo Code as an MCP server to an MCP host.

Check that it works

This is not a connectable MCP server but an AI coding agent platform. After installing the CLI, run `kilo` in any project directory; if the interactive session starts, the install worked.

Troubleshooting

  1. Confirm you downloaded the correct release asset for your platform (Windows, macOS Apple Silicon/Intel, Linux x64/ARM); use the x64-baseline build on older CPUs without AVX, and the musl build in glibc-less containers like Alpine
  2. Make sure you didn't install the `.vsix` (VS Code extension package) when you intended to install the CLI binary
  3. Model access requires creating an account first; if model calls fail, verify account setup and model selection
  4. If an automated pipeline behaves unexpectedly, check whether `--auto` was used and is skipping confirmations you expected

Use cases

Generating and editing code from natural language across multiple files, in VS Code, JetBrains, or the terminal
Speeding up coding with inline ghost-text autocomplete suggestions
Using the built-in Plan, Ask, Debug, and Review agents for architecture design, codebase Q&A, troubleshooting, and change review
Running unattended automation in CI/CD with `kilo run --auto` (e.g., running tests and fixing failures)
Connecting external MCP servers via the built-in MCP marketplace to extend what the agent can do

Overview

Kilo Code is an open source coding agent that works as a VS Code extension, JetBrains plugin, CLI tool, or cloud agent. It supports 500+ AI models with mid-task switching, bills at the model provider's rate with no markup, and requires no API keys to get started. The README describes a built-in "MCP marketplace" for finding and wiring up external MCP servers to extend the agent's capabilities, but there is no evidence in the source material that Kilo Code itself exposes MCP tools or acts as an MCP server that other clients connect to. Fields specific to being a connectable MCP server (transport, tool list, client compatibility, install config for an MCP host) are left empty accordingly.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 64.1k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 080ca6c41399 Data synced 2026-11-04 Read the FMRS scoring method