Skip to main content

Quick Start

apistash exposes a single MCP endpoint at https://api.apistash.io/mcp. Point any MCP-compatible client at that URL — with an API key or OAuth token — and your agent reaches the tools you picked for it. Connecting does not require a separate installer; persistent task-start setup is optional and user-invoked.

1. Choose the tools your agent gets​

Sign in at app.apistash.io and open Settings → MCP Settings. Tick the tools you want your agents to have, and save.

This step is deliberate: an agent's tool list is sent to the model on every session, so a catalogue of everything would cost context and make the model's tool choice worse. You pick a handful, and that is exactly what your agents see. You can change the selection at any time — connected clients pick it up without reconnecting.

Not sure what to pick?

Start with the tools you can name a use for — calculator, datetime_now, and uuid_generate_v4 are a good first set — and add more as you need them. The authenticated system tools setup, bootstrap, and agent_context are always available and do not appear in this selection. If you use agent definitions, grant Agents too. Browse the full catalogue for what is on offer.

2. Connect​

Generate an API key in the dashboard. When creating it, configure the governed categories the key may reach — at least Tools for the calculator in this walkthrough, plus Agents if you assign an agent definition. Tools and Agents are independent. With no category selected, only the three authenticated system tools are available over MCP; no governed tool, resource, prompt, or agent definition is exposed. Then add the following values through your API-key client's native MCP configuration. The exact file and field names vary, so use MCP Clients for the client-specific shape. Claude Desktop instead uses an OAuth custom connector described there; do not put this API-key example in claude_desktop_config.json.

{
	"mcpServers": {
		"apistash": {
			"type": "http",
			"url": "https://api.apistash.io/mcp",
			"headers": {
				"X-API-Key": "YOUR_API_KEY"
			}
		}
	}
}

That's it. Restart your client and the tools you selected will appear.

Prefer OAuth?

apistash publishes standard discovery metadata, so an OAuth-capable client can walk you through sign-in after you register that client in the dashboard. apistash does not dynamically register an unknown client. See Authentication.

3. Verify it's working​

Ask your AI assistant to use one of the tools you selected — for example, if you enabled the calculator:

"Use the calculator tool to evaluate sqrt(2) * pi."

If the tool call succeeds and returns a result, you are connected.

Connected, but no tools?

If the three system tools appear but the tool you chose does not, two usual causes are an empty selection under Settings → MCP Settings or a key without the Tools capability. Check the key's detail page. See Troubleshooting.

4. Optional: set up task-start context​

If you want your client to try apistash/bootstrap at the start of later tasks, explicitly ask your agent to set it up:

"Set up apistash."

General setup also offers an optional conversation to propose useful Agents, Prompts, and Resources. The agent checks existing content and shows concrete drafts and ownership targets for confirmation before governed writes. You can decline; repair-only requests do not start the interview.

The agent calls apistash/setup, then uses the native persistent instruction mechanism supported by your client. It asks before writing if the right target or scope is ambiguous, preserves unrelated instructions, and reports the outcome. If the client cannot save persistent instructions, it shows you the canonical instruction instead and reports that setup was not installed. See the complete task-start setup guide for each supported client's native project scope and the next-task selection flow.

Next steps​