One MCP connection that provisions everything your agent needs — model routing, on-demand GPUs, live search, document processing, memory and media generation. Plug in. It's handled.
# one connection, every resource $ claude mcp add --transport http agentilityx https://mcp.agentilityx.ai \ --header "Authorization: Bearer $AGX_KEY" ✓ connected — 4 tools available > route_llm("summarize this contract", task="write") [lane: nemotron-3-super · 1,204 tokens · $0.0009] Done.
Each service is one tool call, metered per request. Start with routing today; the rest of the catalog rolls out from stacks we already run in production.
Text and vision tasks answered by the best available model lane, with automatic failover and per-request metering.
Quote and launch the right rented GPU per job, billed by the second. Weights cached close to the metal, loaded in seconds, shut down when idle.
Fresh web and news search with resolved sources, so agents answer with facts — reverse-image lookup included.
Bulk OCR, PDF to clean text, content-level dedup, and a searchable library built from any document pile. Proven on a million-file legal corpus.
Images, video clips, talking avatars, speech synthesis and transcription — commercial-safe models, one tool call each.
Persistent memory across sessions and private semantic-search collections, created with one call and owned by you.
Same socket, two doors.
You have an agent or a workflow and want results: route a task, generate media, search the live web, process documents. Pay per request or a flat monthly plan — no infrastructure to think about.
Resellers and product teams get the raw infrastructure: metered API keys you re-issue to your customers, white-label endpoints, volume GPU rates and a revenue-share partner program. Your product, your brand, our provisioning.
Built MCP-native from day one: your agent speaks to one endpoint, and the platform does the provisioning dance behind it.
Add the AgentilityX endpoint to any MCP-capable agent — Claude Code, desktop clients, or your own stack — with a single bearer key.
The router classifies your request, picks the cheapest capable lane or spins up the right GPU, loads cached weights, and runs the job.
Everything unloads when idle. Every request lands in one metering ledger — per-second GPU billing, transparent usage, no idle charges.
Works with any MCP client over streamable HTTP. Examples below use Claude Code and plain curl.
Request early access below — keys arrive by email with a starter allowance.
claude mcp add --transport http agentilityx https://mcp.agentilityx.ai \ --header "Authorization: Bearer YOUR_KEY"
curl -X POST https://mcp.agentilityx.ai/ \
-H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
"name":"route_llm",
"arguments":{"prompt":"three bullet points on MCP","task":"write"}}}'
route_llm — route a text or vision task to the best lane and get the answer.
gpu_quote — which GPU a task class needs and what it costs per hour.
gpu_run — launch a metered GPU job (gated during early access).
ump_status — your usage and spend, last 24 hours.
Early access while we finish the billing portal — generous allowances, founder pricing locked in for early customers.