Connect Your Agent

Skilder acts as an MCP server. Your AI agent connects to it the same way it connects to any MCP server — over Streamable HTTP, secured by MCP OAuth 2.1. Once connected, the agent automatically discovers the capabilities (skills and tools) available through its assigned roles.

The Connect page in Skilder has everything you need. Click Connect from the top navigation bar.

Step 1: Copy the link

On the Connect page, the Any client card at the top shows one link:

https://your-skilder-instance/mcp

This is the same link for every client and every user — it doesn't carry a key. Copy it.

Step 2: Add it to your client

Paste the link into your client's MCP server settings. Most clients call this "Add an MCP server" or "Add a connector." A few concrete examples:

Claude Desktop / Claude.ai / Claude Code

Add the link as a remote MCP server. Claude Code: claude mcp add --transport http skilder https://your-skilder-instance/mcp.

Cursor, VS Code (Copilot), ChatGPT, Microsoft Copilot Studio

Each has its own "Add MCP server" flow in settings — paste the link there. See the client's own card on the Connect page for a link to its MCP docs.

Any other MCP client

The Set up a client section on the Connect page lists more (Windsurf, Cline, Slack, NanoClaw, Openclaw, Hermes, Antigravity CLI...). Not listed? Use the same link anyway — every MCP client connects the same way. Once it connects, it shows up automatically under Detected on the Connect page, so you can find it again by name.

The first time the client uses the link, it opens a one-time browser sign-in to Skilder. Approve it, and the connection is live — nothing to authorize from inside Skilder itself.

Step 3: Test the connection

Send a message to your agent. It should discover and use the skills available through its roles. Try:

What capabilities do you have available?

The agent should list the skills and their tools.

If it doesn't work, check:

• The sign-in actually completed (some clients silently fail the redirect — try disconnecting and reconnecting)
• The role has skills assigned with tools linked
• MCP servers are in Connected status (check Workspace Settings > MCP Servers)

How the MCP OAuth connection works

Your Agent ←→ MCP OAuth + Streamable HTTP ←→ Skilder ←→ MCP Servers (tools)

1. Your agent adds the Skilder link as a remote MCP server.
2. On first use, the client itself redirects through Skilder's OAuth consent screen — this is a normal browser sign-in, not a button inside Skilder's UI.
3. Skilder issues that client its own scoped access. The agent discovers capabilities through MCP's standard tool listing.
4. When the agent calls a tool, Skilder routes the request to the right MCP server, executes it, and returns the result.

Your agent never talks directly to the underlying MCP servers — Skilder handles routing, permissions, and execution.

Advanced: clients that can't do MCP OAuth yet

Every client page has a collapsed Advanced section for the cases OAuth doesn't cover:

Run locally (stdio) — configure your client to run the Skilder runtime directly, authenticated by an API key:

{
  "command": "npx",
  "args": ["-y", "@skilder-ai/runtime"],
  "env": {
    "USER_KEY": "your_key_here"
  }
}

Use a key instead of OAuth — for remote clients that only support a static key, append it to the link:

https://your-skilder-instance/mcp?key=your_key_here

Create an API key from the same Advanced section if you don't have one yet.

Using Studio Instead

If you want to test without an external client, use the built-in Studio. Click Studio in the top navigation — it connects directly to your workspace's capabilities with no extra setup. You just need an AI provider configured.

Next Steps

• Quickstart — Full walkthrough of building your first skill and connecting.
• API Keys — Understand key types, scoping, and best practices.
• Studio — Test skills interactively without an external client.
• Troubleshooting — Common connection issues and fixes.