SmartGate MCP server
2026년 11월 2일 커밋알고리즘 게이트웨이를 통해 AI agent의 리서치, 컨텍스트 압축, 메모리 관리 및 토큰 예산 집행을 지원하는 인텔리전스 레이어입니다.
다른 이름 smartgate-network/smartgate-workflows, smartgate-workflows
- 스타
- 3
- Tools
- 12
- 다운로드, 30일
- —
- 최신 버전
- 1.0.0
- 마지막 커밋
- 2026년 11월 2일
- 라이선스
- MIT
SmartGate MCP란?
SmartGate은(는) MCP server입니다. 알고리즘 게이트웨이를 통해 AI agent의 리서치, 컨텍스트 압축, 메모리 관리 및 토큰 예산 집행을 지원하는 인텔리전스 레이어입니다. 12개 도구를 제공합니다. 예: smart_search, smart_fetch, smart_dedup.
프로젝트 README에서 생성된 요약
SmartGate MCP server 정보
SmartGate는 AI agent와 외부 데이터 소스 사이에서 작동하도록 설계된 인텔리전스 레이어입니다. 리서치, 컨텍스트 압축 및 메모리 관리를 처리하여 agent가 외부 세계와 상호작용하는 방식을 관리하는 특화된 게이트웨이를 제공합니다. 챗봇이나 모델 라우터로 작동하는 대신, 토큰 예산을 강제함으로써 무분별한 도구 사용으로 인한 예상치 못한 비용 발생을 방지하는 알고리즘 레이어 역할을 수행합니다.
개발자는 SmartGate 대시보드에서 발급받은 API key를 사용하여 streamable HTTP를 통해 이 서비스에 연결할 수 있습니다. Cursor, Claude Code, claude.ai, Zed를 비롯하여 Goose 또는 OpenClaw와 같은 다양한 MCP 지원 환경과 호환됩니다. 이 시스템은 smart_search 및 smart_memory와 같은 7가지 특정 프리미티브를 사용하여 단일 인터페이스를 통해 복잡한 작업을 실행합니다.
이 게이트웨이를 활용하면 agent는 소스 검색, 페이지를 Markdown으로 읽기, 호스트 모델에 전달하기 전 정보 중복 제거와 같은 다단계 작업을 수행할 수 있습니다. 이를 통해 컨텍스트를 깨끗하게 유지하고 정의된 제한 범위 내로 관리할 수 있어, 신뢰할 수 있는 리서치 및 메모리 기능이 필요한 자율형 agent를 구축하는 개발자에게 필수적인 도구입니다.
활용 사례
- 1자동 웹 리서치agent가 여러 온라인 소스를 찾아 페이지 콘텐츠를 Markdown으로 변환하여 처리를 용이하게 할 수 있습니다.
- 2토큰 예산 관리개발자가 agent에게 스스로 토큰 사용량을 모니터링하고 제한하도록 지시하여 높은 비용 발생을 방지할 수 있습니다.
- 3지속적인 팀 지식 관리서로 다른 세션 간에 기억해야 할 특정 사실을 저장하고 검색하도록 AI에게 작업을 맡길 수 있습니다.
- 4컨텍스트 최적화agent가 중복 정보를 제거하고 텍스트를 압축하여 모델로 전송되는 데이터 양을 줄일 수 있습니다.
프로젝트의 README를 바탕으로 작성되었습니다. 모든 활용 사례는 해당 내용의 인용문과 대조하여 확인합니다.
SmartGate MCP 설치 방법
형식 확인: 2026년 11월 10일Cursor에 추가Goose에 추가Claude Code에 SmartGate MCP 설치
$ claude mcp add --transport http smartgate-workflows https://smartgate.network/api/mcpCursor에 SmartGate MCP 설치
~/.cursor/mcp.json
{
"mcpServers": {
"smartgate-workflows": {
"url": "https://smartgate.network/api/mcp"
}
}
}VS Code에 SmartGate MCP 설치
.vscode/mcp.json
{
"servers": {
"smartgate-workflows": {
"type": "http",
"url": "https://smartgate.network/api/mcp"
}
}
}VS Code CLI에 SmartGate MCP 설치
$ code --add-mcp '{"name":"smartgate-workflows","type":"http","url":"https://smartgate.network/api/mcp"}'Codex에 SmartGate MCP 설치
$ codex mcp add smartgate-workflows --url https://smartgate.network/api/mcpCodex config.toml에 SmartGate MCP 설치
~/.codex/config.toml
[mcp_servers.smartgate-workflows] url = "https://smartgate.network/api/mcp"
Gemini CLI에 SmartGate MCP 설치
$ gemini mcp add -s user --transport http smartgate-workflows https://smartgate.network/api/mcpGemini settings.json에 SmartGate MCP 설치
~/.gemini/settings.json
{
"mcpServers": {
"smartgate-workflows": {
"url": "https://smartgate.network/api/mcp",
"type": "http"
}
}
}SmartGate MCP 도구 12
README에서 감지됨smart_searchsmart_fetchsmart_dedupsmart_context_gatesmart_budget_guardsmart_memorysmart_pipesmartgate-workflowssmartgate-research-agentsmartgate-deep-readingsmartgate-remember-and-recallsmartgate-usage-report
| smart_search | — | |
| smart_fetch | — | |
| smart_dedup | — | |
| smart_context_gate | — | |
| smart_budget_guard | — | |
| smart_memory | — | |
| smart_pipe | — | |
| smartgate-workflows | — | |
| smartgate-research-agent | — | |
| smartgate-deep-reading | — | |
| smartgate-remember-and-recall | — | |
| smartgate-usage-report | — |
자동 검사
2026년 11월 14일에 실행됨전체 보고서 →아직 인증되지 않았습니다. 서버는 80점 이상의 점수, 심각한 결함 없음, 그리고 검사에 충분한 데이터를 갖추어야 합니다. 검사 방식 · 점수를 높이기 위해 수정할 항목
- 소스 저장소 연결 및 접근 가능smartgate-network/smartgate-workflows is public and reachablegithub.com/smartgate-network/smartgate-workflows ↗통과
- 패키지가 이 저장소를 소스로 명시검사되지 않음
- 빌드 출처(Sigstore 증명)검사되지 않음
- 오픈 소스 라이선스 (OSI)MITspdx.org/licenses/MIT.html ↗통과
- 최근 1년 내 커밋 또는 릴리스Last commit or release 2026-11-02github.com/smartgate-network/smartgate-workflows ↗통과
- OpenSSF Scorecard검사되지 않음
- 패키지 이름이 인기 패키지와 유사하지 않음검사되지 않음
- 패키지 및 해당 종속성에 중요한 보안 권고 없음검사되지 않음
- 모델에 대한 숨겨진 지침 없음12 tool descriptions and the README read; no override phrases, credential paths or hidden characters통과
- 파괴적인 도구에 라벨 표시됨No tool named like delete, drop or exec통과
- 원격 엔드포인트가 MCP 핸드셰이크에 응답HTTPS endpoint answered 401 (sign-in required)통과
README
GitHub에서 보기SmartGate
AI Intelligence Layer — not another assistant or LLM proxy.
Official SmartGate plugin for Cursor, Claude Code, claude.ai, and Cowork.
SmartGate is the intelligence layer between AI and the world — helping you research, understand, remember, and act before runaway tool traffic becomes a surprise invoice.
On the surface: Smart Research, decisions, knowledge, and monitoring — no MCP required. Under the hood: an MCP-native algorithmic gateway with five capabilities (Research, Context, Memory, Control, Pipeline) powered by seven smart_* primitives. Human interface (Web) + agent interface (MCP/API). Not a chatbot, search app, or model router. The host model stays on the client.
Plugin mark for Cursor Marketplace is the 1:1 shield assets/logo.png. The banner is the official site OG card. The research demo is the same GIF as the homepage.
What is SmartGate MCP
Don't rebuild research infrastructure for every agent. Connect once over Streamable HTTP; the gateway fetches, searches, compresses, deduplicates, remembers, and hard-caps tool traffic before it reaches your model provider.
| Capability | What the agent does | Product |
|---|---|---|
| Research | Find sources, then read pages as Markdown | Traffic Customs |
| Context | Dedup overlapping hits, then compress | Token Control |
| Memory | Persist and recall team facts across sessions | Features |
| Control | Check, count, and cap token budget | Token Control |
| Pipeline | Run research / read / remember as one job | Tools |
- Product: smartgate.network · Features · Pricing
- Connect: Quick start · Connect · MCP endpoint
- Reference: Tools · Security · llms.txt
- Compliance: Audit · Privacy · Terms
Install
Mint a key at Dashboard → API Keys (sk_live_…). Never paste the key into chat or commit it.
Official MCP Registry name: network.smartgate/smartgate. Same endpoint: https://smartgate.network/api/mcp. The listing is metadata only — you still need the Dashboard Bearer key.
Cursor
- Install SmartGate from the Cursor Marketplace, or import this repo under Cursor Dashboard → Plugins → Team Marketplaces → Import from Repo:
https://github.com/smartgate-network/smartgate-workflows - Open Plugins → SmartGate → Configure and set
SMARTGATE_API_KEY - Restart Cursor (or reload the window) so the MCP handshake runs again
Claude (claude.ai / Cowork / Claude Code)
- Add SmartGate from Anthropic's directory, or load this repo locally:
claude --plugin-dir . - When Claude prompts for plugin configuration, set
SMARTGATE_API_KEY. In Claude Code you can also run/plugin configure @smartgate - Confirm the connector: on claude.ai / Cowork use the plugin's Connectors tab; in Claude Code run
/mcpand look forplugin:smartgate:smartgate
Gemini CLI
gemini extensions install https://github.com/smartgate-network/smartgate-workflows
Set SMARTGATE_API_KEY when the CLI prompts. Manifest: gemini-extension.json.
OpenClaw
Install the ClawHub skill, then add the hosted MCP:
openclaw skills install @smartgate.network/smartgate
Connect: Settings → MCP → Add server (Streamable HTTP, https://smartgate.network/api/mcp, Authorization: Bearer ${SMARTGATE_API_KEY}). Skill source: skills/smartgate/SKILL.md.
Hermes
From the Hermes MCP catalog:
hermes mcp install smartgate
Paste the Dashboard sk_live_… key when prompted (MCP_SMARTGATE_API_KEY).
Or add the hosted server in ~/.hermes/config.yaml and the key in ~/.hermes/.env:
mcp_servers:
smartgate:
url: "https://smartgate.network/api/mcp"
headers:
Authorization: "Bearer ${SMARTGATE_API_KEY}"
X-SmartGate-Agent-Platform: hermes
Skill (ClawHub):
hermes skills search smartgate --source clawhub
hermes skills install clawhub/smartgate.network/smartgate
Or add this repo as a skill tap: hermes skills tap add smartgate-network/smartgate-workflows.
AutoGPT
From AutoGPT's hosted MCP catalog, select SmartGate and paste the Dashboard sk_live_… key as the access token. Paste the key only — the client adds Bearer.
Or add an MCP Tool block: URL https://smartgate.network/api/mcp, auth Bearer, access token = Dashboard API key.
Zed
Settings → AI → MCP Servers → Add Server → Add Remote Server:
{
"context_servers": {
"smartgate": {
"url": "https://smartgate.network/api/mcp",
"headers": {
"Authorization": "Bearer ${SMARTGATE_API_KEY}",
"X-SmartGate-Agent-Platform": "zed"
}
}
}
}
Or install from Zed MCP extensions if SmartGate is listed there.
VS Code / GitHub Copilot
Command Palette → MCP: Add Server → HTTP. URL https://smartgate.network/api/mcp. Set header Authorization = Bearer ${SMARTGATE_API_KEY}.
Or browse the gallery: Extensions view → search @mcp smartgate.
OpenCode
Add a remote MCP in OpenCode config:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"servers": {
"smartgate": {
"type": "remote",
"url": "https://smartgate.network/api/mcp",
"oauth": false,
"headers": {
"Authorization": "Bearer {env:SMARTGATE_API_KEY}",
"X-SmartGate-Agent-Platform": "opencode"
}
}
}
}
}
Goose
Add a Streamable HTTP extension: URL https://smartgate.network/api/mcp, header Authorization: Bearer ${SMARTGATE_API_KEY}, X-SmartGate-Agent-Platform: goose. Skill source: skills/smartgate/SKILL.md.
Kilo Code
Install from the Kilo Marketplace, or connect the hosted MCP the same way as Cline — see llms-install.md.
Crush
Add a remote Streamable HTTP MCP in Crush config. URL https://smartgate.network/api/mcp, header Authorization: Bearer ${SMARTGATE_API_KEY}.
Cherry Studio
Settings → MCP → MCP Servers → Add. Import from JSON (type must be streamableHttp) or Quick Create → Streamable HTTP:
{
"mcpServers": {
"smartgate": {
"type": "streamableHttp",
"url": "https://smartgate.network/api/mcp",
"headers": {
"Authorization": "Bearer ${SMARTGATE_API_KEY}"
}
}
}
}
Enable the server, then Work → Agent → Edit → MCP and bind it.
AnythingLLM
Settings → Agent Skills, or edit plugins/anythingllm_mcp_servers.json (type is streamable, not streamable-http):
{
"mcpServers": {
"smartgate": {
"type": "streamable",
"url": "https://smartgate.network/api/mcp",
"headers": {
"Authorization": "Bearer ${SMARTGATE_API_KEY}"
}
}
}
}
Reload MCP servers in the UI, start SmartGate, then enable its tools for the workspace.
Other MCP clients (LibreChat, Open WebUI, Dify, Langflow, Flowise, n8n)
Any host that supports remote Streamable HTTP can connect with:
- URL:
https://smartgate.network/api/mcp - Header:
Authorization: Bearer ${SMARTGATE_API_KEY} - Official Registry name:
network.smartgate/smartgate
First run
After the MCP server shows as connected, try:
Check my SmartGate token allowance with smart_budget_guard action=check.
smart_fetch https://smartgate.network/docs/tools then summarize Token Control, Traffic Customs, and Audit.
Remember: our staging API rate limit is 100 req/min
Live tools/list schemas (purpose, params, returns) come from the hosted server, not this repo. Names are smart_* — there is no smartgate_compress_context. Full reference: Docs → Tools.
Tools (reviewer cheat sheet)
Every tool returns a JSON string. Parse it before synthesizing. Do not invent page content.
| Tool | Required | Optional | Returns |
|---|---|---|---|
smart_search |
query |
max_results (1–50, default 10) |
Search hits (results[].url, snippet) |
smart_fetch |
url (http/https) |
timeout (default 30s) |
Page markdown, title, url |
smart_dedup |
texts (string[]) |
threshold (0–1, default 0.9) |
Kept / removed indices or deduped texts |
smart_context_gate |
text |
ratio (default 0.5), purpose |
compressed, ratio, token counts |
smart_budget_guard |
action: check | count | record |
text / tokens by action |
Allowance, usage, or record ack |
smart_memory |
action: add | search | get | delete |
text (add), query (search), memory_id (get/delete) |
Shape depends on action |
smart_pipe |
template or steps |
query, url, text |
Per-step results. Templates: research, read, remember. Not for a single step. |
Skills shipped with this plugin:
| Skill | Routes |
|---|---|
smartgate-workflows |
Pick the right workflow |
smartgate-research-agent |
Multi-source research |
smartgate-deep-reading |
One known URL |
smartgate-remember-and-recall |
Team memory |
smartgate-usage-report |
Quota / usage (no content) |
Configure
Cursor Marketplace: Plugins → SmartGate → Configure → SMARTGATE_API_KEY. Cursor reads mcp.json (${SMARTGATE_API_KEY}).
Claude directory / Claude Code: /plugin configure @smartgate → SMARTGATE_API_KEY. Claude reads .mcp.json (${user_config.SMARTGATE_API_KEY}). Do not commit a real key.
Manual / reviewer fallback (only if you are not using the plugin): merge this into ~/.cursor/mcp.json (user) or <project>/.cursor/mcp.json (project), then restart Cursor. Replace the placeholder. If the plugin is also installed, delete this block so tools are not duplicated.
{
"mcpServers": {
"smartgate": {
"type": "streamable-http",
"url": "https://smartgate.network/api/mcp",
"headers": {
"Authorization": "Bearer sk_live_REPLACE_WITH_YOUR_KEY",
"X-SmartGate-Agent-Platform": "cursor"
}
}
}
}
Claude Code / claude.ai fallback (only if the plugin is not installed). Same endpoint, Claude's http type, and ${user_config.SMARTGATE_API_KEY} in the plugin copy of .mcp.json:
{
"mcpServers": {
"smartgate": {
"type": "http",
"url": "https://smartgate.network/api/mcp",
"headers": {
"Authorization": "Bearer sk_live_REPLACE_WITH_YOUR_KEY",
"X-SmartGate-Agent-Platform": "claude"
}
}
}
}
Hosted endpoint: POST https://smartgate.network/api/mcp (Streamable HTTP). Mint a key at Dashboard → API Keys. If you also add the standalone SmartGate MCP connector from the directory, use the same URL so tools are not duplicated.
Data handling
This plugin does not run fetch, search, compression, or memory inside Cursor or Claude. Those steps run on the hosted gateway at https://smartgate.network/api/mcp. The host model (Cursor or Claude) stays on the client.
- In transit: HTTPS +
Authorization: Bearer. The plugin ships no secrets. You setSMARTGATE_API_KEYin the host's plugin configure dialog (Cursor variables, or ClaudeuserConfig). - At rest: we store a hash of API keys, not the raw secret. Teams are isolated by the key's tenant. Do not pass a foreign
team_id. - What is logged: tool name, timestamp, token counts, and for fetch the requested URL. Retention: Free 7 days · Pro 30 · Teams 90 · Enterprise 180. See Security and Logs.
- Compression / dedup:
smart_context_gate(LLMLingua) andsmart_dedup(SemHash) run on SmartGate servers, not locally. They shrink or drop overlapping text; they are not a PII redaction product. Do not put secrets or credentials in tool arguments or memory. - What we do not do: we are not an LLM proxy and do not train a public chatbot on your prompts as a consumer product. Billing card data is handled by Paddle, not stored by us.
Privacy Policy · Terms · SECURITY.md · support@smartgate.network
Troubleshooting
| Symptom | Check |
|---|---|
| 401 Unauthorized | SMARTGATE_API_KEY is set in the host's plugin configure dialog; key is sk_live_… |
| Tools empty | Restart Cursor, or in Claude Code run /mcp and reconnect plugin:smartgate:smartgate |
| 405 on GET | Endpoint is POST /api/mcp, not GET SSE |
| Duplicate tools | Remove a hand-written smartgate block from ~/.cursor/mcp.json or ~/.claude.json |
Workflow source
workflows/ is the scenario source (YAML + Markdown). skills/ is the committed plugin pack, generated by:
pip install jsonschema pyyaml
python scripts/install-skills.py --client cursor --target skills --force
python tests/validate.py
| Workflow | Description |
|---|---|
| research-agent | search → fetch → dedup → compress |
| deep-reading | fetch → compress |
| remember-and-recall | team memory |
| usage-report | budget check / count |
Manual skill install (no Marketplace): see clients/README.md.
Maintainers
SmartGate · support@smartgate.network
License
SmartGate MCP 질문
SmartGate MCP를 Claude Code에 추가하려면 어떻게 하나요?
터미널에서 claude mcp add --transport http smartgate-workflows https://smartgate.network/api/mcp를 실행합니다. 모든 프로젝트에서 사용할 수 있게 하려면 --scope user를 추가합니다.
SmartGate MCP를 Cursor에 추가하려면 어떻게 하나요?
~/.cursor/mcp.json를 열고 mcpServers 아래에 서버를 추가합니다. 위 설치 섹션에 표시된 Cursor 구성을 사용합니다.
SmartGate MCP는 어떤 도구를 제공하나요?
12개 도구를 제공합니다. 예: smart_search, smart_fetch, smart_dedup, smart_context_gate, smart_budget_guard.
SmartGate는 LLM 프록시인가요, 아니면 챗봇인가요?
아니요, SmartGate는 LLM 프록시, 챗봇, 검색 앱 또는 모델 라우터가 아닙니다. 호스트 모델은 클라이언트에 그대로 유지됩니다.
MCP 게이트웨이는 어떤 기능을 제공하나요?
게이트웨이는 Research, Context, Memory, Control, Pipeline의 다섯 가지 핵심 기능을 제공합니다.