gem-hunter
Language: TypeScript
Open-source AI CLI and local MCP server connecting 25 clients—Claude Code, Cursor, Codex, ChatGPT, Hermes, and OpenClaw—to 2,000+ models/APIs, with OAuth and rollback.
Give your AI agent superpowers. One line. 2,000+ AI models and APIs. 100 free data API calls a day, then pay per call from one balance.
English | 中文 | 日本語 | 한국어 | Español | Français | Deutsch | Português
Your AI coding assistant is smart, but it's trapped in a box. It can't search the web, check social media, generate images, or access real-time data — unless you wire up each API yourself.
SandBase changes that. One command connects your agent to 2,000+ AI models and APIs through the Model Context Protocol. No API keys to manage. No configuration headaches.
Building directly over HTTP instead of MCP? Start with the SandBase API quickstart or compare the unified LLM, image, and video API with specialized model gateways.
Paste this into your agent and let it do the rest:
set up https://www.sandbase.ai/SKILL.md
It reads the SandBase Skill, runs connect itself, and asks you to approve one browser sign-in. Prefer to drive it yourself:
npx -y @sandbaseai/cli@latest connect
Or install the official Homebrew formula on macOS or Linux:
brew install sandbaseai/tap/sandbaseai-cli sandbase connect
See the Homebrew launch announcement and macOS/Linux CI evidence.
Approve the browser authorization once; each supported client can then discover,
inspect, and run available SandBase models and APIs through the local MCP bridge.
Start a new session afterwards so the sandbase_* tools load.
Every GitHub release publishes its own SHA-256 next to the tarball, so you can pin
and verify an exact build instead of tracking latest:
V=$(npm view @sandbaseai/cli version) curl -fL -O "https://github.com/sandbaseai/cli/releases/download/v$V/sandbaseai-cli-$V.tgz" shasum -a 256 "sandbaseai-cli-$V.tgz" # compare against the checksum on the release page npx -y "./sandbaseai-cli-$V.tgz" connect
Want to inspect compatibility before signing in or changing configuration? The client catalog is a read-only command — it writes nothing and needs no account:
npx -y @sandbaseai/cli@latest catalog --json
A no-write catalog audit and a bounded doctor check against an empty isolated profile.
Homebrew users can run the same read-only check with sandbase catalog --json.
| Price | |
|---|---|
| Data APIs — search, social, scraping, e-commerce, finance | 100 free calls/day per organization, then paid per call (price shown by sandbase inspect) |
| Discovery, inspection, run status, balance, file upload | Free |
| Models — LLM, image, video, audio | Paid per call or per million tokens |
sandbase inspect <tool> always shows the current price before you run anything;
data APIs covered by the daily allowance also show pricing.free_quota. The 100
free calls are shared across eligible data APIs and reset at 00:00 UTC; after
that, calls are billed per call. sandbase balance --json shows today's remaining free
calls as free_api_quota. Prices can change, so trust the live response over
any example in this file.
Visit the official SandBase CLI project page for a compact overview.
For a security-focused walkthrough of client detection, MCP tools, and rollback,
read the 25-client SandBase CLI setup guide.
AI agents and automated installers can use the concise
llms-install.md instructions.
main.Discover the authenticated remote bridge in the official MCP Registry,
read its Registry community showcase, or find it through
Chinese Independent Developer,
Awesome MCP,
Glama,
TensorBlock MCP Index,
MCPRepository,
VaultPlane,
AIMCP,
MCP Server Hub,
MCP Server Spot, or the
Awesome AI API Proxy.
External maintainers have merged SandBase into
Awesome Codex CLI,
Awesome Agent Skills, and
Awesome MCP Server.
The Hermes integration is independently listed in the
Hermes Atlas ecosystem catalog and as beta in
Awesome Hermes Agent.
Install its
verified sandbase Agent Skill from skills.sh,
the sandbaseai/cli/sandbase listing on SkillsCat,
the sandbaseai/cli/sandbase listing on skills.re,
the 93/100 verified SandBase listing on vSkill,
the security-audited SandBase listing on Skillstore,
or the versioned sandbaseai/sandbase snapshot on Agent Skill Hub.
After connecting, ask your agent to follow the same inspectable sequence for search, data APIs, language models, image generation, video generation, and other catalog capabilities:
sandbase_discover to find candidates by task or capability.sandbase_inspect to read the current input schema, pricing, and
execution requirements.sandbase_run.
Data APIs include 100 free calls per organization per UTC day, then are
billed per call like models; confirm the cost shown by sandbase_inspect
before paid runs.sandbase_run_get with the returned run_id
instead of creating a duplicate run.sandbase_runs to review recent run status and recorded cost.sandbase_account to check the current account balance.For example, start with a non-billable discovery request:
Find image models suitable for a square product illustration. Compare the required inputs and current pricing for the best two candidates. Do not run either model yet.
Discovery and inspection never cost anything, so an agent can always survey the catalog before it spends.
After reviewing the live response, explicitly approve the endpoint and inputs you want to run. Catalog entries, schemas, pricing, latency, and availability can change, so use the tool response from the current session rather than a static example.
See the complete Claude Code and Codex walkthrough or the Chinese workflow guide. For a live catalog example, read the GitHub connector discover-and-inspect dogfooding note.
| Category | Tools | What You Can Do |
|---|---|---|
| Web Search | Google, Exa, Tavily, Scholar | Research anything in real-time |
| Social Media | Twitter/X, YouTube, Reddit, Instagram, TikTok, Xiaohongshu, Weibo, Bilibili | Monitor trends, pull posts, analyze content |
| Scraping | Firecrawl, Exa Content | Extract data from any public webpage |
| AI Models | GPT-4o, Claude, Gemini, DeepSeek, Qwen, Llama | Run any model without managing keys |
| Image Gen | Flux, DALL-E, Ideogram, Recraft | Create logos, illustrations, mockups |
| Video Gen | Kling, MiniMax, Runway, Luma | Generate short clips and animations |
| Audio | ElevenLabs, Whisper | Text-to-speech, transcription |
| E-commerce | Taobao, Amazon | Product research, price comparison |
| Finance | Stock data, company info | Market research and analysis |
Works with every major AI coding tool:
| Auto-configured | Manual setup |
|---|---|
| Cursor, Cursor CLI | ChatGPT |
| Claude Code | Claude Desktop |
| Codex | |
| Kiro IDE, Kiro CLI | |
| Windsurf | |
| Gemini CLI | |
| Amp | |
| Warp | |
| OpenCode, Qwen Code, Kimi CLI | |
| Hermes, OpenClaw |
Preview the complete compatibility catalog without signing in or changing any files:
npx -y @sandbaseai/cli@latest catalog --json
# Connect all detected clients at once npx -y @sandbaseai/cli@latest connect # Or target one specific client npx -y @sandbaseai/cli@latest connect --client cursor
Target OpenClaw directly, then ask the CLI to read back the managed MCP server and SandBase Skill before you rely on them:
npx -y @sandbaseai/cli@latest connect --client openclaw npx -y @sandbaseai/cli@latest doctor --client openclaw
After doctor reports the credential, MCP bridge, and Skill as configured,
restart OpenClaw and make a natural-language request that needs SandBase. If
OpenClaw already has a non-SandBase MCP server named sandbase, the installer
leaves it untouched instead of overwriting it. To remove the managed MCP entry
and local credential later:
npx -y @sandbaseai/cli@latest unregister --client openclaw
OpenClaw currently exposes no supported Skill removal command, so unregister
reports that limitation and leaves the verified SandBase Skill installed.
SandBase can install both its managed Hermes Skill and local MCP entry in one authorization transaction:
npx -y @sandbaseai/cli@latest connect --client hermes npx -y @sandbaseai/cli@latest doctor --client hermes
doctor verifies the owned local credential, MCP entry, and Skill. Restart
Hermes to confirm native discovery in a fresh client session, then use the
SandBase Skill or MCP tools. A pre-existing, non-SandBase MCP entry named
sandbase is left untouched. To remove only the managed Hermes configuration,
Skill, and local credential while retaining the shared bridge for other clients:
npx -y @sandbaseai/cli@latest unregister --client hermes
The integration is independently listed by Hermes Atlas and Awesome Hermes Agent.
┌─────────────────┐ ┌───────────────┐ ┌────────────────────┐
│ Your Agent │ MCP │ Local Bridge │ HTTPS │ SandBase API │
│ (Cursor, etc.) │────────▶│ (on-demand) │────────▶│ 2,000+ models/APIs │
│ │◀────────│ │◀────────│ │
└─────────────────┘ └───────────────┘ └────────────────────┘
connect → CLI opens browser auth → you approve0600)sandbase connect [--client <name>] # Authorize + configure sandbase setup # After upgrading: refresh the local bridge + Skill, test connectivity sandbase doctor [--client <name>] # Local checks plus one real round trip to SandBase sandbase unregister [--client <name>] # Remove configuration sandbase catalog --json # List all supported clients
Agents without MCP, or with a broken MCP session, use the same catalog from a shell. Credentials come from connect automatically (or SANDBASE_API_KEY / sandbase keys add -k <key>).
sandbase discover -q "twitter posts" # search; -t api|llm|multimodal|embedding|service sandbase inspect sandbase_twitter_web_search_timeline # fields, price, ready-to-run command sandbase run sandbase_twitter_web_search_timeline -i '{"keyword":"AI agents"}' sandbase run <video_tool> -i '{"prompt":"..."}' -w -o result.json # wait with backoff, save output to disk sandbase runs get -r <run_id> -w -o result.json # resume a long run instead of re-running it sandbase upload ./photo.png # local image/audio/video -> public URL for tools that take a URL sandbase runs list | sandbase balance
-j prints JSON; errors are {"error":{"code","message","hint"}} with the next step in hint. Exit codes: 0 success or still running, 1 failure (including a failed run), 2 usage.
Sync tools (LLMs, most data APIs) return their result directly; async tools (image, video, audio jobs) return a run id — wait with -w, or fetch later with sandbase runs get -r <run_id>. If api.sandbase.ai is unreachable, the bridge keeps the SandBase tools visible with an actionable error instead of dropping the server.
| Tool | Purpose |
|---|---|
sandbase_discover |
Search all 2,000+ available AI models and APIs — free |
sandbase_inspect |
Get input schema, current price, and a ready-to-run template — free |
sandbase_run |
Execute a model or API endpoint — eligible data APIs: 100 free calls/day per organization, then paid per call; models: paid per call |
sandbase_run_get |
Poll status/result of an async run (video gen, etc.) — free |
sandbase_runs |
List your recent runs with their cost — free |
sandbase_account |
Check account balance and today's free data API calls (free_api_quota) — free |
sandbase_run_cancel |
Cancel an Agent Service run — free |
Recommended workflow:
discover → inspect → run
1. sandbase_discover(q: "twitter search") → find current candidates
2. sandbase_inspect(name: discovered_name) → get schema and pricing
3. sandbase_run(name: discovered_name, input: {...}) → run after confirmation
0600unregister, only configuration owned by SandBase is removednpx -y @sandbaseai/cli@latest connect
Then ask your agent to do something it couldn't before.
Create an account → — every organization gets 100 free data API calls a day.
If SandBase saves you setup time, star the repository — it helps more agent users discover the project.
Community-curated listings:
Apache-2.0