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

OpenClaude Improved

Community
Runs anywhere, uses anything
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #391 of 438
Stars
★ 569 Very popular
Transport
stdio (local process)
Runtime
Node.js >=22 · Bun 1.3.13+
Credentials
Optional API key
License
Other / unspecified
Last commit
Tools
10
35FMRS · D

OpenClaude Improved is a powerful CLI coding agent supporting multiple LLM providers, retaining many core features of Claude Code while adding capabilities like provider routing, MCP support, and more. It is suitable for developers who need model flexibility, but be aware of its unofficial status and the risks of executing commands.

Strongest · Maintenance 12/20 Weakest · Reliability 3/20

Reliability
3/20
Security and permissions
4/20
Maintenance
12/20
Documentation
10/20
Setup experience
6/20
Why each score
Reliability 3/20
The repository root is the server path, but no MCP server manifest (e.g., mcp-server configuration or MCP-related entries in package.json) was found. The README describes a CLI coding agent, not explicitly an MCP server. Core code and tests are claimed, but no verifiable execution evidence is present; the only evidence is documentation of build and test commands. Per static calibration, reliability cannot exceed 12, and committed tests covering key paths are absent, so a low score is warranted.
Security and permissions 4/20
Security score is low due to significant unmitigated risks: the project directly allows arbitrary tool execution (Bash, read/write/edit) without explicit confirmation or sandboxing. The README mentions external providers like Opengateway and credentials stored in a local config file, but does not clarify encryption or least-privilege principles. No malware or credential theft is evident, but unconfirmed dangerous operations and insufficient credential handling details justify a score of 4.
Maintenance 12/20
The repository is active with 569 stars and 3 open issues, not archived. Licenses are MIT (though derived Claude Code code remains Anthropic's). Build and test scripts indicate ongoing maintenance. However, there is no release history or versioning information, and no details on a security response channel (SECURITY.md is mentioned but not included in source). Due to missing release and governance evidence, score is 12.
Documentation 10/20
The README covers installation, quick start, providers, configuration, and gotchas. However, there is no MCP-server-specific protocol documentation, tool parameters, limits, or cost details. No troubleshooting section (unless gotchas count). Docs are usable but assume familiarity with Claude Code environment. Lack of verifiable evidence for install or usage. Score is 10.
Setup experience 6/20
Install steps are clear (Node >=22 and Bun required, then build and install globally). However, there are no example client configurations for mainstream MCP clients like Claude Desktop, and no specific MCP server configuration examples. Platform compatibility is stated (Windows/macOS/Linux), but no verified step-by-step guide. Due to lack of verifiable execution evidence, setup score is capped at 15 by static calibration, here 6.

Static review · not runListed 2026-08-07

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 to use non-Anthropic models but keep Claude Code-style workflows
  • Multi-provider environments needing support for various cloud APIs and local models
  • Those who want advanced agent features and tool loops in the CLI
  • Users needing flexible provider/model configuration

Not for

  • Users requiring official Anthropic certification or enterprise support
  • Non-technical users who need a graphical interface
  • Users expecting zero-configuration out-of-the-box experience
  • Those needing official Claude Code compatibility layer (this is an independent fork, not official)

Required permissions

  • File system access: read, write, edit code files
  • Execute system commands (Bash)
  • Network access: call cloud APIs (OpenAI, Gemini) or local Ollama service
  • May access API keys from environment variables
  • Read/write access to user config directory (~/.openclaude/)

Risks and side effects

  • Running arbitrary bash commands may cause system damage or data loss, especially with untrusted prompts
  • API keys may be exposed through environment variables; handle carefully
  • Project is under active development; APIs may change
  • Non-official fork; potential compatibility or stability issues
  • MCP tools may introduce third-party code execution risk

Setup

Before you start

Runtime：Node.js >=22 · Bun 1.3.13+

OPENAI_API_KEY optionalsecret API key for OpenAI or any compatible gateway; save via /provider or pass as env var.
FIRECRAWL_API_KEY optionalsecret Optional; enables JS-rendered web fetching; get it from firecrawl.dev.
AIMLAPI_API_KEY optionalsecret Key for AI/ML API (api.aimlapi.com/v1).
FIREWORKS_API_KEY optionalsecret Key for Fireworks AI, 276 curated models.
LONGCAT_API_KEY optionalsecret Key for LongCat (api.longcat.chat/openai/v1).
CLINE_API_KEY optionalsecret Key for ClinePass (api.cline.bot/api/v1), with usage caps.
OPENCODE_API_KEY optionalsecret Key for OpenCode Zen/Go (opencode.ai).
MIMO_API_KEY optionalsecret Key for Xiaomi MiMo (mimo.mi.com), uses api-key header auth.
NEARAI_API_KEY optionalsecret Key for NEAR AI (cloud-api.near.ai/v1).
CLOUDFLARE_API_TOKEN optionalsecret API token for Cloudflare Workers AI; get it from the Cloudflare dashboard.
Other optional settings (6)
CLAUDE_CODE_USE_OPENAI optional Set to 1 to enable the OpenAI-compatible backend; export from your shell.
OPENAI_MODEL optional Model name, e.g. gpt-4o or qwen2.5-coder:7b for Ollama.
OPENAI_BASE_URL optional Points to an OpenAI-compatible /v1 endpoint, e.g. http://localhost:11434/v1 for Ollama.
OPENCLAUDE_OLLAMA_NUM_CTX optional Overrides the Ollama context window (32768 requested by default).
OLLAMA_CONTEXT_LENGTH optional Alternative Ollama-side context length override.
OPENCLAUDE_CONFIG_DIR optional Relocates the config directory from the default ~/.openclaude/.
  1. Ensure Node.js >= 22 and Bun are installed.
  2. Clone the repository and navigate into it.
  3. Run bun install to install dependencies.
  4. Run bun run build to build the project.
  5. Run npm install -g . to install the CLI globally.
  6. Run openclaude to start the agent.

Check that it works

After building and globally installing, run the openclaude command; if the interactive CLI starts and the read or bash tool can answer a question about the current directory, the install works. If grep reports 'ripgrep not found', install ripgrep system-wide.

Troubleshooting

  1. 'ripgrep not found' on startup: install ripgrep system-wide and ensure `rg --version` works
  2. Ollama connection issues: ensure Ollama is running on the default port (11434) and adjust `OPENCLAUDE_OLLAMA_NUM_CTX` or `OLLAMA_CONTEXT_LENGTH`
  3. Provider API authentication errors: check environment variables are set correctly, use `/provider` to configure interactively
  4. Build failures: ensure Node.js version meets requirements, rerun `bun install`
  5. Session anomalies: use `--resume` or `--fork-session` to recover

Things to try

Once connected, you can ask your AI assistant things like:

  • Use the bash tool to list all files in the current directory and count them
  • Read src/index.ts and summarize what it does
  • Search the repo for TODO comments with the grep tool and list where they are
  • Rewrite the install steps from the README and write them into SETUP.md

Tools 10

bash writes
Execute shell commands locally
read read-only
Read file contents
write writes
Write file contents
edit writes
Edit file contents
grep read-only
Search text within files
glob read-only
Match file paths using glob patterns
agents writes
Spawn sub-agents for specific tasks
tasks writes
Create and manage background tasks
Show 2 more tools
mcp writes
Integrate with external tools via MCP protocol
slash commands read-only
Execute built-in slash commands like /provider and /buddy

Use cases

AI-assisted coding with any supported provider
Developing with local models like Ollama
Extending coding workflows with MCP tools
Using headless mode for CI
Interactive chat in VS Code

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

OpenClaude Improved is an open-source CLI coding agent built from Claude Code, but designed to work with many LLM providers, including OpenAI-compatible APIs (like OpenRouter, DeepSeek, Groq), Ollama local models, Gemini, GitHub Models, and more. It provides the same tools and agent workflows as Claude Code, including bash execution, file read/write/edit, search, multi-step tool loops, streaming output, vision support, repo mapping, and more. It also includes advanced features like agent routing, MCP support, a gRPC server, and a VS Code extension.

Similar servers

Context7 80 · B

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

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 645538c4fd9d Data synced 2026-08-10 Read the FMRS scoring method