Potencia tu agente
El comando propio del publicador. No es un endpoint de PlugMCP.
Host a copy with our keys and logs.
npx -y @codespar/[email protected]Claude Code no tiene enlace de instalación para un comando local. Copia el comando de abajo y ejecútalo en una terminal.
claude mcp add codespar-sdk -- npx -y @codespar/sdk@0.16.29Ejecútalo en una terminal.
{ "mcpServers": { "codespar-sdk": { "command": "npx", "args": [ "-y", "@codespar/[email protected]" ] } }}claude_desktop_config.json
¿URL remota o un comando en tu máquina? En qué se diferencian las dos instalaciones
Ejecutar en PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirDel publicador
Commerce infrastructure for AI agents. Payments, invoicing, shipping, notifications, and ERP across Latin America — one SDK.
LATAM-first by design: Pix + NF-e + WhatsApp + PSP routing are first-class, with MCP servers for the rails (Zoop, Nuvem Fiscal, Melhor Envio, Z-API, Omie, and more) curated in codespar/mcp-dev-latam. Any framework, any agent — channel-agnostic by design.
Commerce SDK for AI agents — sessions, managed auth, Complete Loop orchestration for Latin American APIs
Core
Framework adapters — convert session tools to your framework's tool format
npm install @codespar/sdk # 0.10 line
import { CodeSpar } from "@codespar/sdk";
const cs = new CodeSpar({ apiKey: "ak_..." });
// Create a session with the Brazilian commerce preset.
// `preset: "brazilian"` resolves to a curated set of MCP servers
// covering Pix charges (Zoop), NF-e issuance (Nuvem Fiscal), shipping
// labels (Melhor Envio), WhatsApp (Z-API), and ERP (Omie). Swap the
// preset or pass `servers: [...]` explicitly to customize.
const session = await cs.create("user_123", {
preset: "brazilian",
manageConnections: { waitForConnections: true },
// projectId: "prj_a1b2c3d4e5f6g7h8", // optional — defaults to the org's default project
});
// Natural language
const result = await session.send(
"Charge R$150 via Pix and issue the NF-e"
);
// Or execute tools directly
const charge = await session.execute("asaas/create_payment", {
customer: "cus_xxx",
billingType: "PIX",
value: 150,
});
// Don't know which tool to call? Discover by use case via the meta-tool router.
// Returns recommended + related, plus connection status so the agent knows
// whether to set anything up first.
const found = await session.discover("send a pix payment");
console.log(found.recommended);
// { server_id, tool_name, connection_status, known_pitfalls, ... }
// Surface the connect wizard when a needed server is disconnected.
// Returns a deep-link the UI renders into a Connect button — the agent
// never receives credentials directly.
const wizard = await session.connectionWizard({
action: "initiate",
server_id: found.recommended!.server_id,
});
console.log(wizard.initiate?.connect_url);
// Typed meta-tool wrappers — same wire as session.execute("codespar_charge", ...)
// but you get a typed payload back instead of the generic ToolResult envelope.
const charge = await session.charge({
amount: 150,
currency: "BRL",
method: "pix",
buyer: { name: "Cliente Demo", document: "11144477735" },
});
const label = await session.ship({
action: "label",
origin: { /* ... */ },
destination: { /* ... */ },
items: [/* ... */],
});
// Double-entry ledger (codespar_ledger) — routes to your self-hosted
// Midaz instance. Amounts in minor units; source debits must equal
// destination credits.
const posting = await session.ledger({
action: "entry",
asset: "BRL",
source: [{ account: "@external/BRL", amount: 12500 }],
destination: [{ account: "@wallet/user_1", amount: 12500 }],
});
// Async settlement — codespar_charge / codespar_pay return synchronously,
// but real settlement lands via webhook. Poll, or stream over SSE.
const settled = await session.paymentStatus(charge.tool_call_id);
await session.paymentStatusStream(charge.tool_call_id, {
onUpdate: (env) => console.log(env.status), // pending → succeeded / failed / refunded
});
// Async KYC — codespar_kyc returns the inquiry id; the buyer finishes
// the hosted flow off-platform. Poll or stream the disposition.
const v = await session.verificationStatus(inquiryToolCallId);
await session.verificationStatusStream(inquiryToolCallId, {
onUpdate: (env) => console.log(env.status), // pending → approved / rejected / review / expired
});
// Complete Loop — full commerce workflow (meta-tool driven)
import { loop } from "@codespar/sdk";
const result = await loop(session, {
steps: [
{ tool: "codespar_charge", params: { amount: 150, currency: "BRL", method: "pix", buyer: { name, document: cpf } } },
{ tool: "codespar_invoice", params: (prev) => ({ rail: "nfe", company_id, payment_id: prev[0].data.id, /* ... */ }) },
{ tool: "codespar_ship", params: { action: "label", origin, destination, items } },
{ tool: "codespar_notify", params: { recipient: phone, message: "Order received." } },
],
onStepComplete: (step, r) => console.log(`✓ ${step.tool}: ${r.duration}ms`),
});
Beyond direct canonical-tool calls, the platform registers 14 commerce-grade meta-tools that route to the best provider per request, with failover, idempotency, and per-tenant connection config. They split into a buy side (the agent spends), a sell side (the agent sells), and shared infrastructure:
Buy side
Sell side
Shared
Use session.execute("codespar_pay", { amount, currency, recipient })
or the typed wrappers (session.discover, session.connectionWizard).
The router picks the best provider per call based on the tenant's
connections + cost/latency telemetry. See docs/tool-router.md
for how the router selects a provider per call.
Buy-side spending is governed by a signed mandate that backs a unified
multi-currency wallet: one signature mints per-currency slots (for
example BRL via Pix plus USDC), each with its own total cap and per-tx
cap. There is no FX inside the wallet; the payee routes the spend to the
matching currency slot, and a cross-currency move is a real ramp trade
at the real rate. From @codespar/cli:
codespar mandate create --slot BRL:pix:50000:10000 --slot USDC:usdc:100:25
codespar wallet <consumer> # rolled up per currency
codespar transfer <consumer> --from BRL --to USDC --amount 15000
Each adapter exposes getTools(session) returning tools shaped for that framework. Same session, any framework — pick one or compose several.
// Vercel AI SDK
import { getTools } from "@codespar/vercel";
const tools = getTools(session);
const result = await generateText({ model: openai("gpt-4o"), tools, prompt: "..." });
// Claude Agent SDK
import { getTools } from "@codespar/claude";
const tools = getTools(session);
// OpenAI Agents SDK
import { getTools, handleToolCall } from "@codespar/openai";
const tools = getTools(session);
// LangChain.js / LlamaIndex.TS / Mastra / CrewAI / AutoGen / CAMEL / Letta / Gemini
import { getTools } from "@codespar/langchain"; // or @codespar/llama-index, @codespar/mastra, etc.
const tools = getTools(session);
// MCP (Claude Desktop, Cursor, VS Code)
import { getMcpConfig } from "@codespar/mcp";
const { url, headers } = getMcpConfig(session);
Projects let you run independent environments — dev, staging, prod — inside a single org, each with its own connections, usage, and audit trail. Pin a client to a project with new CodeSpar({ apiKey, projectId: "prj_..." }), or override per call via cs.create(userId, { projectId: "prj_..." }). Session-level projectId wins over client-level; omit both and requests fall back to the org's default project.
When set, the SDK sends an x-codespar-project header on session creation and on all MCP transport calls. See docs.codespar.dev/concepts/projects for the full managed-tier concept.
npm install
npm run build # Build all packages
npm run test # Run all tests
npm run typecheck # Type check all packages
For production workloads with governance, audit trails, policy engines, self-hosted runtimes, and enterprise commerce primitives (mandates, payment routing), see CodeSpar Enterprise.
MIT — codespar.dev
Commerce SDK for AI agents — sessions, managed auth, Complete Loop orchestration for Latin American APIs
{ "mcpServers": { "codespar-sdk": { "command": "npx", "args": [ "-y", "@codespar/[email protected]" ] } }}Herramientas
Aún no hay lista de herramientas. Los servidores remotos la reciben tras la primera comprobación de estado.
Preguntas frecuentes
Copia en esta página la configuración stdio para Cursor, Claude Desktop, Claude Code o VS Code. Usa la instalación del publicador, no un proxy de PlugMCP.
Este registro no indica ninguna variable de entorno obligatoria.
@codespar/sdk se ejecuta localmente por stdio.
Última comprobación: 2026-11-13. Commerce SDK for AI agents — sessions, managed auth, Complete Loop orchestration for Latin American APIs