1. Conduid
  2. Browser
  3. Engine
MCP server · Browser

Engine

WeaveKit engine — a metadata-driven headless backend runtime with Git-versioned metadata

Unclaimed other
55Fair

Scored 22 hours ago · breakdown

About Engine

Engine is an MCP server in the Browser category: weaveKit engine — a metadata-driven headless backend runtime with Git-versioned metadata.

The repository has 2 stars and 0 forks, with the last commit yesterday. Recent commit activity is the strongest signal that reported issues will get attention.

Install

Claude Code
claude mcp add engine -- npx -y @weave-kit/engine
npx
npx -y @weave-kit/engine

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Tools (14)

adaptersList supported providers

Parameters not published.

askAsk ClawAPI anything

Parameters not published.

audio_transcribeTranscribe audio files

Parameters not published.

embeddingsGenerate text embeddings

Parameters not published.

growth_guideGrowth guide: progress, recommendations, pool health

Parameters not published.

image_generateGenerate images

Parameters not published.

keys_addAdd a new API key

Parameters not published.

keys_listView your API keys

Parameters not published.

llmChat with any AI model through ClawAPI

Parameters not published.

searchWeb search via Brave/Tavily/DuckDuckGo

Parameters not published.

Show all 14 tools
setup_wizardFirst-time setup: scan env for keys, validate, Claw Key

Parameters not published.

statusCheck engine health

Parameters not published.

taskExecute multi-step AI tasks

Parameters not published.

translateTranslate text via DeepL or AI

Parameters not published.

Ask AI

✦

Ask AI about Engine

Powered by Claude · Grounded in docs

I know everything about Engine. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

中文版 · 日本語版


✨ Why ClawAPI

What happens How
① Zero-config setup Scans your env for API keys, validates, imports — done in 10 seconds setup_wizard auto
② Smart recommendations After setup, tells you exactly which free provider to add next growth_guide recommend
③ Never stuck on rate limits Groq quota hit? Auto-switches to Gemini. Tells you how to double your quota L2 Smart Gateway
④ Gets smarter over time Anonymous usage data improves routing for everyone Collective Intelligence

One command to manage all your AI keys. One engine that routes every request to the best provider.


The Problem

You have API keys scattered across OpenAI, Anthropic, Google, DeepSeek, Groq...

  • Keys stored in .env files across 20 projects
  • No idea which key is burning money
  • Can't easily switch providers when one goes down
  • AI coding tools (Claude Code, Cursor) each need separate key config

The Solution

         ┌─────────────────────────────────────────────┐
         │              ClawAPI Engine                  │
         │           (runs on YOUR machine)             │
         │                                              │
  You ──►│  🔑 Encrypted key vault (AES-256-GCM)       │
         │  🧠 Smart routing across providers           │
         │  📊 Cost tracking & health monitoring        │
         │  🔌 OpenAI-compatible API on localhost       │
         │                                              │
         │   Keys     Keys     Keys     Keys     Keys  │
         │    │        │        │        │        │     │
         └────┼────────┼────────┼────────┼────────┼─────┘
              ▼        ▼        ▼        ▼        ▼
           OpenAI  Anthropic  Gemini  DeepSeek  Groq
                                              + 10 more

Your keys never leave your machine. Period.


⚡ Quick Start

Install via npm (requires Bun)

# Install
bun add -g @clawapi/engine

# Setup (interactive — adds your first API key)
clawapi setup

# Start the engine
clawapi start

Or download a binary (no dependencies)

# macOS (Apple Silicon)
curl -fsSL https://github.com/sstklen/clawapi/releases/latest/download/clawapi-darwin-arm64 -o clawapi
chmod +x clawapi && ./clawapi setup
Platform Download
macOS Apple Silicon clawapi-darwin-arm64
macOS Intel clawapi-darwin-x64
Linux x64 clawapi-linux-x64
Windows x64 clawapi-win-x64.exe

→ All releases


🔌 Use with AI Coding Tools

Claude Code (MCP) — Recommended

Prerequisites: Bun or Node.js 20+ · Claude Code installed

Step 1: Add ClawAPI to Claude Code

claude mcp add clawapi --scope user -- bunx @clawapi/engine mcp

Step 2: Restart Claude Code (close and reopen your terminal)

Step 3: Verify it works

clawapi mcp --test

You should see: ✅ MCP Server OK with tool count and engine status.

Where is the config stored? Claude Code saves MCP config at ~/.claude.json. You can check it with cat ~/.claude.json.

Quick setup (optional): Generate default config without interactive prompts:

clawapi setup --defaults

You now have 14 AI tools available. Ask Claude: "What tools do you have from ClawAPI?"

Tool What it does
llm Chat with any AI model through ClawAPI
search Web search via Brave/Tavily/DuckDuckGo
translate Translate text via DeepL or AI
image_generate Generate images
audio_transcribe Transcribe audio files
embeddings Generate text embeddings
keys_list View your API keys
keys_add Add a new API key
status Check engine health
adapters List supported providers
setup_wizard First-time setup: scan env for keys, validate, Claw Key
growth_guide Growth guide: progress, recommendations, pool health
ask Ask ClawAPI anything
task Execute multi-step AI tasks

Any OpenAI SDK Client

from openai import OpenAI

# Point any OpenAI client at ClawAPI — it just works
client = OpenAI(
    base_url="http://localhost:4141/v1",
    api_key="your-clawapi-key"
)

# ClawAPI picks the best available provider automatically
response = client.chat.completions.create(
    model="auto",  # Let ClawAPI choose, or specify "gpt-4" / "claude-3" / "gemini-2"
    messages=[{"role": "user", "content": "Hello!"}]
)

Works with: Python, Node.js, Go, Rust — anything that speaks OpenAI API.


🧠 Smart Routing (L1 → L4)

ClawAPI doesn't just proxy — it thinks.

Layer Name What it does
L1 Direct Proxy Fastest path. Pass request to the specified provider.
L2 Smart Gateway Auto-select the best provider based on cost, latency, and health.
L3 AI Concierge Understand intent, pick the right model + parameters.
L4 Task Engine Break complex tasks into steps, orchestrate multiple AI calls.
"Translate this doc to Japanese and summarize it"

  L4 Task Engine
   ├─ Step 1: L1 → DeepL (translate)
   ├─ Step 2: L2 → Best LLM (summarize)
   └─ Step 3: Merge results → Return

🔑 Iron Rules

These are not features — they are guarantees.

# Rule How
1 Keys never leave your machine All API calls made locally. VPS only sees metadata.
2 VPS never sees API content ECDH P-256 key exchange. Only latency/status shared.
3 Works offline Full functionality without internet. VPS is optional.

📦 Supported Providers

Provider Models Type
OpenAI GPT-4o, GPT-4, o1, o3 LLM
Anthropic Claude 4, Claude 3.5 Sonnet LLM
Google Gemini 2.5, Gemini 2.0 Flash LLM
DeepSeek DeepSeek-V3, DeepSeek-R1 LLM
Groq Llama 3, Mixtral (ultra-fast) LLM
Cerebras Llama 3 (fastest inference) LLM
SambaNova Llama 3 (fast inference) LLM
OpenRouter 200+ models (aggregator) LLM
Qwen Qwen-2.5 LLM
Ollama Any local model LLM
Brave Search Web search Search
Tavily AI-powered search Search
DuckDuckGo Web search (free) Search
DeepL 30+ languages Translation
+ Community adapters (YAML) Extensible

Add your own provider in 30 lines of YAML. No code needed.


🛠 Full CLI

Engine      start · stop · status
Keys        keys add · list · remove · pin · rotate · import · check
Claw Key    claw-key set · show · remove
Sub-Keys    sub-keys issue · list · revoke · usage
Mutual Aid  aid config · stats · donate
Adapters    adapters list · install · remove · update
Telemetry   telemetry show · toggle
Backup      backup export · import
System      logs · config · setup · doctor · version · mcp

30+ commands. 3 languages (English, 繁體中文, 日本語).


🏗 Architecture

┌─────────────────────────────┐          ┌────────────────────────┐
│      ClawAPI Engine         │          │     ClawAPI VPS        │
│      (Your Machine)         │  ECDH    │     (Optional Cloud)   │
│                             │◄────────►│                        │
│  🔐 Key Vault (AES-256)    │ Metadata │  📋 Device Registry    │
│  🧠 Smart Router (L1-L4)   │   Only   │  📊 Telemetry Agg.    │
│  🌐 OpenAI-Compat API      │          │  🤝 Mutual Aid Match  │
│  🔧 MCP Server (14 tools)  │          │  🔍 Anomaly Detection │
│  💻 CLI (30+ commands)      │          │                        │
│  🖥  Web UI (SSR + HTMX)   │          │                        │
└─────────────────────────────┘          └────────────────────────┘
      Keys stay here ☝️                    Never sees your keys

🔒 Security

  • AES-256-GCM encryption at rest
  • ECDH P-256 key exchange with VPS
  • 1,681 tests, 0 failures
  • Triple code review (self + Codex + Opus cross-review)
  • 5-party security audit methodology
  • Non-root Docker execution
  • Rate limiting on all endpoints

Tech Stack

Component Technology
Runtime Bun
Framework Hono
Database SQLite (bun:sqlite)
Language TypeScript
Packaging Bun compile (4-platform binaries)
Container Docker + Caddy

🩺 Included: Dr. Claw — AI Debug Knowledge Base

This repo includes Dr. Claw — an AI debugging knowledge base that remembers every bug ever solved. Ask once, never repeat.

  • 6,800+ issues indexed, 980+ verified fixes
  • MCP Server + REST API
  • Spun off from Confucius Debug, now lives in clawapi/drclaw/

📝 License

AGPL-3.0 — Free to use, modify, and distribute. Contributions welcome.

See LICENSE for details.

Pair With

  • Opus Relay (sstklen/opus-relay) — Bridge your local Claude CLI to any VPS. ClawAPI manages the keys, Opus Relay bridges the compute.

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About Engine

How do I install Engine?

Run claude mcp add engine -- npx -y @weave-kit/engine, then add the server to your MCP client's configuration.

Is Engine safe to use with an AI agent?

Its trust score is 55 out of 100 (fair). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Engine still maintained?

Yes — the last commit was yesterday. The repository has 2 stars and 0 forks.