Integrations

GCTRL connects to any MCP client, agent framework, or system that can make an HTTP call. There is no proprietary protocol to adopt - pick the transport that matches where your agent runs. Remote agents only need one thing: the web origin (port 3001, your TLS domain, or a tailnet hostname) - see Networking & Ports. Full marketplace view: gctrl.tech/integrations.

Connection methods

MCP over HTTP (remote agents)

{
  "mcpServers": {
    "gctrl": {
      "type": "http",
      "url": "https://<your-install>/api/agent/mcp",
      "headers": { "Authorization": "ApiKey <token>" }
    }
  }
}

Off by default - enable the gateway in Settings → Agent. No separate port to open: the gateway is reachable at /api/agent/mcp on the same origin as the app UI.

MCP over stdio (local agents)

{
  "mcpServers": {
    "gctrl": {
      "command": "node",
      "args": ["services/mcp/dist/index.js"],
      "env": { "GCTRL_API_TOKEN": "gctrl_your_scoped_token" }
    }
  }
}

Build once with cd services/mcp && npm install && npm run build.

Direct HTTP (any framework, zero MCP)

curl -X POST https://<your-install>/api/agent/tools/search_entities \
  -H "Authorization: ApiKey <token>" \
  -H "Content-Type: application/json" \
  -d '{ "query": "Ada Lovelace" }'

Every tool GCTRL exposes over MCP is also callable directly at POST /api/agent/tools/<tool> - useful for frameworks with their own tool-calling convention. GET /api/agent/skill.md returns the canonical GCTRL memory skill as markdown, for frameworks that take a drop-in system-prompt file instead of an MCP config.

Every integration sees exactly the clearance and knowledge-base scope carried by its token - create tokens in Access Control.

After connecting: install the skill

Whichever transport you pick, the MCP config only gives the agent the tools. Right after connecting, install the GCTRL skill too - it teaches the agent when to read which memory layer and to always write conclusions back, which is what makes GCTRL's memory compound across sessions instead of sitting there unused. Canonical copy: gctrl.tech/skill.md.

Coding agents

Agent	Method
Claude Code	MCP http config in `.mcp.json`
Codex	MCP config (`~/.codex/config.toml`)
Kimi	MCP
Cline	MCP
OpenClaw	MCP or direct HTTP tools

IDEs

IDE	Method
Cursor	MCP (`.cursor/mcp.json`)
Windsurf	MCP
Zed	MCP
VS Code + GitHub Copilot	MCP support / agent mode

Copilots

Copilot	Method
Microsoft Copilot Studio	Custom connector → GCTRL REST API, `Authorization: ApiKey` header
GitHub Copilot	MCP in agent mode

Agent frameworks

Framework	Method
Pi (built-in GCTRL agent)	Zero setup - enable in Settings → Agent
Paperclip	Drop-in: copy `skill.md` (`GET /api/agent/skill.md`), call `POST /api/agent/tools/<tool>`
Hermes	Same drop-in harness pattern as Paperclip
LangChain	HTTP tools (`POST /api/agent/tools/<tool>`) or the MCP adapter
LlamaIndex	Same - HTTP tools or MCP adapter

Automation

Tool	Method
n8n	Native GCTRL nodes - see Settings → n8n
Webhooks	Settings → Webhooks

Knowledge sources

Source	Method
Google Drive	Connector
Obsidian	Connector
SharePoint	Connector
Website crawler	Connector
PDF / DOCX upload	Upload in the UI, or `ingest_file` from any connected agent