Sandbox
Two curls and about a minute. The first mints a throwaway key, the second runs a genuine HTTP call through the reliability engine. No account, no card, nothing to uninstall afterwards.
No signup. The key is disposable and rate limited per IP, so grab one and move on.
$ curl -sX POST https://api.sandbox.klanexai.com/playground/key
{
"api_key": "klx_test_9f2k7c1e4b8a…",
"webhook_secret": "whsec_test_5d3a…",
"base_url": "https://api.sandbox.klanexai.com",
"mcp_url": "https://api.sandbox.klanexai.com/mcp",
"included_executions": 100,
"expires_at": "2026-08-18T09:14:22Z"
}
A real POST to a real host. klanex validates the payload against your schema, queues it, and hands back an execution ID in milliseconds.
$ curl -s https://api.sandbox.klanexai.com/v1/executions \
-H "X-API-Key: klx_test_9f2k7c1e4b8a…" \
-H "Content-Type: application/json" \
-d '{
"target": { "url": "https://httpbingo.org/post" },
"payload": { "charge_id": "ch_9f2k", "amount": 500 },
"payload_schema": {
"type": "object",
"required": ["charge_id", "amount"],
"properties": { "amount": { "type": "integer" } }
},
"idempotency_key": "demo-ch_9f2k"
}'
{"execution_id": "exe_4d0c85a6", "status": "QUEUED"}
Every attempt is on the execution document, queryable for as long as your retention window allows. Run it immediately after step 2 and you may still catch QUEUED or RUNNING; give it a second and read it again.
$ curl -s https://api.sandbox.klanexai.com/v1/executions/exe_4d0c85a6 \
-H "X-API-Key: klx_test_9f2k7c1e4b8a…"
{
"execution_id": "exe_4d0c85a6",
"status": "SUCCEEDED",
"target": { "method": "POST", "url": "https://httpbingo.org/post" },
"attempts": 1,
"max_attempts": 5,
"result": { "status_code": 200, "body": "{\"args\": {}, \"headers\": {…}, \"method\": \"POST\", \"json\": {\"amount\": 500, \"charge_id\": \"ch_9f2k\"}}" },
"created_at": "2026-08-11T09:14:22Z",
"updated_at": "2026-08-11T09:14:23Z"
}
target.url at
https://httpbingo.org/status/503 and run it again. klanex absorbs the
5xx with exponential backoff, climbs through its attempt budget, and only then
fails the execution terminally with a classified error code. Your agent gets one
answer at the end instead of five failures in its context window.
amount,
or make it a string) and the call is rejected in milliseconds with an
llm_hint written to be pasted straight back into a model's context.
Rejected payloads never count against your quota.
The same sandbox key works over MCP. One command and your agent has the whole engine as native tools.
$ claude mcp add --transport http klanex https://api.sandbox.klanexai.com/mcp \
--header "X-API-Key: klx_test_9f2k7c1e4b8a…"
# your agent can now call, as native tools:
execute get_execution list_executions replay_execution get_usage list_connections
$ KLANEX_API_KEY=klx_test_… npx -y klanex-mcp # stdio-only clients
Works in Claude Code, Claude Desktop, Cursor, VS Code, and Windsurf. Full setup in the MCP docs. One caveat on a sandbox key: replay_execution is a paid-plan feature, so it will come back as a permission error until you upgrade.
A free account is the same engine with 1,000 executions a month, durable history, and no seven-day clock.