One page to connect any AI surface — Claude Code, Claude Desktop, Cursor, Raycast, OpenClaw — to your Synap data pod. Skills, MCP, API keys.

Connect your AI

Give any AI direct access to your Synap data pod. Pick your tool, run one command, done.

Pick your surface

Tool	What you get	Setup	Guide
Claude Code	Agent skills	3 steps, ~2 min	Claude Code →
Claude Desktop	Skills + MCP	3 steps, ~3 min	Claude Desktop →
Cursor	MCP	2 steps, ~2 min	Cursor →
Raycast	Raycast extension with AI tools	3 steps, ~3 min	Raycast →
OpenClaw	Full agent runtime + skills	1 command	OpenClaw →
Other	OpenAI-compat, Hub Protocol, custom	Varies	Custom integration →

The short version

npx @synap-core/cli connect --target=claude-code
# or: --target=claude-desktop, cursor, raycast, openclaw

The CLI:

Asks for your pod URL (defaults to http://localhost:4000 for local pods)
Provisions an API key (or accepts an existing one)
Drops the skills where your tool expects them
Writes the MCP config if the tool supports it
Tells you what to do next

You can skip the CLI and do it manually — every guide has a manual path too.

What "connecting" actually means

Synap exposes three integration primitives. Every AI surface uses one or more:

1. Skills — Markdown files (+ optional scripts) your AI reads to learn how to use Synap. Three skills ship:

synap — core data operations: read, write, link, search, remember
synap-schema — extend the data model (new profiles, new properties)
synap-ui — build views, dashboards, bento layouts

Skills live in your AI's skills directory (~/.claude/skills/ for Claude Code) and load progressively — only when the user asks something the skill is about.

2. MCP — The Model Context Protocol endpoint at {pod}/mcp. Your AI calls it over HTTP/JSON-RPC to invoke Synap tools directly. Authenticated with an API key. Works with Claude Desktop, Cursor, and any MCP-capable client.

3. API key — A bearer token bound to an agent user in your pod. Scoped (hub-protocol.read, hub-protocol.write, mcp.read, mcp.write). Issued via the CLI (synap connect) or manually at POST /api/hub/setup/agent.

Skills + MCP + API key = any AI can use Synap as its sovereign memory.

Governance applies everywhere

Every write through any of the five surfaces above passes through the same proposal layer. Auto-approved actions go straight through; destructive or schema-shaping actions queue for your review.

Auto-approved by default: entity.create, entity.update, relation.create, document.create, view.create, profile.create, property_def.create, memory.*
Always proposal-gated: entity.delete, entity.archive, workspace.create, view.update

Your AI never silently mutates your pod. See governance for the full model.

Don't see your tool?

Three fallback paths:

OpenAI SDK consumers — point OPENAI_BASE_URL at {pod}/v1 and use synap/auto, synap/free, synap/balanced, synap/advanced, or synap/complex as the model. Chat stream guide.
Custom agents — call Hub Protocol REST directly at /api/hub/*. SDK guide.
Automation platforms (n8n, Zapier, Make) — use skill invocation or webhooks. Automation platforms.

Next steps

API keys reference — scopes, rotation, security
Skills architecture — the three skills, files, and progressive disclosure
MCP endpoint reference — protocol details, exposed tools
Governance model — how proposals work