Capabilities

Four callables.
One surface.

Don't hard-code model lists or agent URLs. Discover what's available, inject Skills, attach MCP tools, then call Models and Agents from the same SDK.

50+

Models

Unified inference layer

One API key, 50+ models. OpenAI-compatible completions, embeddings, and vision — routed to the best provider for each request.

  • OpenAI-compatible REST & SSE
  • Weighted fallback routing
  • Per-model budget controls
  • Prompt caching awareness
Explore
v3

Skills

Server-side prompt engine

Version-locked prompt templates + tool schemas injected server-side. Ship a skill ZIP, version-pin with @v3, never ship prompts in client code again.

  • Jinja2 template rendering
  • Version pinning (@v3)
  • Tool schema bundling
  • Immutable after publish
Explore
6-level

MCPs

Model Context Protocol gateway

Proxy any MCP server with full auth (OAuth, PKCE, BYOK) and a 6-level permission cascade. Track per-tool spend, approve or deny access.

  • 6-level permission cascade
  • OAuth2 + PKCE passthrough
  • Per-tool cost tracking
  • URL-namespaced access
Explore
A2A

Agents

A2A runtime

A2A-protocol-compliant agents with agent cards, streaming invocation, and versioned rollback. Register once, discover everywhere.

  • Agent card (.well-known)
  • SSE + NDJSON streaming
  • Versioned snapshots
  • Health check + webhooks
Explore

One discovery endpoint

Everything you can discover via GET /v1/capabilities, you can call. The proxy bundles them transparently \u2014 no client-side orchestration needed.

xct@solar:~
$ curl https://api.xcity.one/v1/capabilities \\
-H "Authorization: Bearer $TOKEN" | jq '.capabilities' | head -40
 
{
  "models": ["gpt-4o", "claude-sonnet-4-6", "deepseek-v3.2", ...],
  "skills": ["fact-check@v3", "code-review@v2", "summarize@v1"],
  "mcps":   ["github", "linear", "jira", "search", "sheets"],
  "agents": ["research-agent", "qa-agent", "deploy-agent"]
}

// pick anything — then call it
$ xct chat completions create \
    --model gpt-4o \
    --skills "fact-check@v3" \
    --tools github.search_issues \
    --message "Find regressions in main"

Try everything today

Free tier includes all four capabilities across the entire model catalog.

Start free