GCTRL
Integrations

Connect your agent. Keep your knowledge.

GCTRL works with any MCP client, agent framework, or direct HTTP call - pick the transport that matches where your agent runs. Every integration sees exactly the clearance and knowledge-base scope carried by its token.

Claude Code

MCP over HTTP via .mcp.json - durable, access-controlled memory in every session.

MCP

Codex

MCP config in ~/.codex/config.toml - same tools, same tokens.

MCP

Kimi

Connects over MCP like any other client.

MCP

Cline

MCP client inside VS Code - point it at the GCTRL gateway.

MCP

OpenClaw

MCP, or call GCTRL tools directly over HTTP.

MCP or HTTP

Cursor

MCP via .cursor/mcp.json.

MCP

Windsurf

MCP client support out of the box.

MCP

Zed

MCP support for agent-mode requests.

MCP

VS Code + GitHub Copilot

Agent mode speaks MCP - same config as any other client.

MCP / agent mode

Microsoft Copilot Studio

Custom connector to the GCTRL REST API - send the ApiKey header, get graph-grounded answers.

Custom connector

GitHub Copilot

Agent mode connects over MCP like any IDE client.

MCP / agent mode

Pi

GCTRL's built-in agent. Zero setup - enable it in Settings → Agent.

Built-in · zero setup

Paperclip

Drop in skill.md and call tools directly over HTTP - no MCP required.

Drop-in · skill.md + HTTP

Hermes

Same drop-in harness pattern as Paperclip.

Drop-in · skill.md + HTTP

LangChain

Wrap GCTRL's HTTP tools, or use the MCP adapter.

HTTP or MCP

LlamaIndex

Same options - HTTP tools or an MCP adapter.

HTTP or MCP

n8n

Native GCTRL nodes - build workflows without touching the API directly.

Native nodes

Webhooks

Fire on ingest, extraction, or fusion events - configure in Settings → Webhooks.

HTTP

Google Drive

Import documents straight from a connected Drive.

Connector

Obsidian

Pull an entire vault in as a knowledge source.

Connector

SharePoint

Import from Microsoft 365 document libraries.

Connector

Website crawler

Point it at a site and ingest the pages that matter.

Connector

PDF / DOCX upload

Upload in the UI, or have any connected agent call ingest_file.

Upload or ingest_file

How it works

Three transports, one set of tools.

Tokens come from Access Control - create one scoped to exactly the knowledge bases an integration should see.

1. MCP over HTTP

Remote agents, orchestrators - the gateway is off by default, enable it in Settings → Agent.

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

2. Install the skill (below) so the agent knows how to use it.

1. MCP over stdio

Local agents - Claude Code, Cursor, Claude Desktop - on the same machine or LAN.

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

2. Install the skill (below) so the agent knows how to use it.

1. Direct HTTP

Any framework with its own tool-calling convention - no MCP client required.

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

2. Install the skill (below) so the agent knows how to use it.

2. Install the GCTRL skill

The config above gives your agent the tools. The skill teaches it how to use them well: run the one-time setup interview, read the right memory layer for each question, and always write conclusions back so GCTRL compounds instead of starting cold every session.

Read the skill docs · gctrl.tech/skill.md

Ready to connect an agent?

Full setup steps for every transport, plus a per-integration reference, live in the docs.

Read the integrations docsQuick Start guide