@openvecta/mcp

MCP server for OpenVecta — call OpenVecta inference models (OpenAI-compatible, USDC-settled on Solana) as tools from any MCP client: Claude Desktop, Cursor, Windsurf, agent kits.

Tools

Tool	What it does	Auth needed
list_models	Catalog + per-1M-token USDC pricing, modality, flat fee	none (public)
estimate_cost	Estimate a call's USDC cost before running it	none (public)
chat	Chat completion on any model	keyed or x402
embed	Embeddings	keyed only

Two auth modes (auto-detected from env)

KEYED — set OPENVECTA_API_KEY (ov_sk_...). Uses your prepaid balance / free tier. Simplest; works everywhere. (Wins if both are set.)

X402 — set SOLANA_PRIVATE_KEY (base58). Pays per call in USDC from that wallet via the x402 protocol — no account, no key. Chat only. ⚠️ Real funds move. Test on devnet first (SOLANA_NETWORK=devnet).

Install

No install step needed — npx fetches it on demand:

npx -y @openvecta/mcp

Add to Claude Desktop / Cursor

Config file — Claude Desktop: %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS).

Keyed:

{
  "mcpServers": {
    "openvecta": {
      "command": "npx",
      "args": ["-y", "@openvecta/mcp"],
      "env": { "OPENVECTA_API_KEY": "ov_sk_live_..." }
    }
  }
}

x402 — pay per call, no account (test on devnet first!):

{
  "mcpServers": {
    "openvecta": {
      "command": "npx",
      "args": ["-y", "@openvecta/mcp"],
      "env": {
        "SOLANA_PRIVATE_KEY": "base58-secret-key",
        "SOLANA_NETWORK": "devnet"
      }
    }
  }
}

Restart the client, then ask e.g. "list the OpenVecta models" or "use OpenVecta's glm-5.2 to explain Solana".

Env reference

Var	Default	Notes
OPENVECTA_API_KEY	—	ov_sk_ key → KEYED mode
SOLANA_PRIVATE_KEY	—	base58 secret → X402 mode
SOLANA_NETWORK	mainnet	mainnet | devnet
SOLANA_RPC_URL	per-network	override RPC
OPENVECTA_BASE_URL	https://api.openvecta.com/v1	staging / self-host

How x402 mode works

Uses the official @x402/fetch client (@x402/svm ExactSvmScheme + a @solana/kit signer) to wrap fetch: the request hits /v1/chat/completions, gets a 402 challenge, the library builds + signs the USDC transfer from your wallet, resends with the payment header, and OpenVecta settles + serves. chat requires max_tokens here (it bounds the pre-authorization ceiling; defaults to 1024 if omitted).

The public model catalogue and per-token pricing are readable without auth at /.well-known/x402.

Security: never commit a private key. For production/agents, load it from a secrets manager, not a config file. list_models / estimate_cost need no auth.

Build from source

npm install
npm run build   # -> dist/index.js

License

MIT