Agent frameworks

@tuppence/agent-tools gives an agent framework the same tools as the MCP server, in that framework’s shape. It fetches them with your key, so the agent sees only what the key allows.

Scroll sideways to see all of it →

1. Install and give it a key

npm install @tuppence/agent-tools
export TUPPENCE_SECRET_KEY=rk_test_…

Use a restricted key with only the scopes the agent needs: for example, payment_links:read and payment_links:write for an agent that sells with links. TUPPENCE_API_KEY also works, or pass apiKey.

2. Hand the tools to your agent

OpenAI Agents SDK

import { Agent, run } from "@openai/agents";
import { tuppenceTools } from "@tuppence/agent-tools/openai-agents";

const { tools, close } = await tuppenceTools();
const agent = new Agent({
  name: "Billing assistant",
  instructions: "Amounts are integers in the currency's minor unit (pence, cents). When you create a payment link, reply with its URL.",
  tools,
});
const result = await run(agent, "Make a £15 payment link for a deep research report.");
console.log(result.finalOutput);
await close();

import { Agent, run } from "@openai/agents";
import { tuppenceTools } from "@tuppence/agent-tools/openai-agents";

const { tools, close } = await tuppenceTools();
const agent = new Agent({
  name: "Billing assistant",
  instructions: "Amounts are integers in the currency's minor unit (pence, cents). When you create a payment link, reply with its URL.",
  tools,
});
const result = await run(agent, "Make a $15 payment link for a deep research report.");
console.log(result.finalOutput);
await close();

import { Agent, run } from "@openai/agents";
import { tuppenceTools } from "@tuppence/agent-tools/openai-agents";

const { tools, close } = await tuppenceTools();
const agent = new Agent({
  name: "Billing assistant",
  instructions: "Amounts are integers in the currency's minor unit (pence, cents). When you create a payment link, reply with its URL.",
  tools,
});
const result = await run(agent, "Make a €15 payment link for a deep research report.");
console.log(result.finalOutput);
await close();

LangChain

import { createAgent } from "langchain";
import { tuppenceTools } from "@tuppence/agent-tools/langchain";

const { tools, close } = await tuppenceTools();
const agent = createAgent({ model: "openai:gpt-5", tools });
const result = await agent.invoke({
  messages: [{ role: "user", content: "Make a £15 payment link for a deep research report." }],
});
await close();

import { createAgent } from "langchain";
import { tuppenceTools } from "@tuppence/agent-tools/langchain";

const { tools, close } = await tuppenceTools();
const agent = createAgent({ model: "openai:gpt-5", tools });
const result = await agent.invoke({
  messages: [{ role: "user", content: "Make a $15 payment link for a deep research report." }],
});
await close();

import { createAgent } from "langchain";
import { tuppenceTools } from "@tuppence/agent-tools/langchain";

const { tools, close } = await tuppenceTools();
const agent = createAgent({ model: "openai:gpt-5", tools });
const result = await agent.invoke({
  messages: [{ role: "user", content: "Make a €15 payment link for a deep research report." }],
});
await close();

Claude Agent SDK

The Claude Agent SDK connects to MCP servers itself, so the package supplies the connection:

import { query } from "@anthropic-ai/claude-agent-sdk";
import { TUPPENCE_TOOLS, tuppenceMcpServer } from "@tuppence/agent-tools/claude-agent-sdk";

for await (const message of query({
  prompt: "Make a £15 payment link for a deep research report.",
  options: {
    mcpServers: { tuppence: tuppenceMcpServer() },
    allowedTools: [TUPPENCE_TOOLS], // or one by one: mcp__tuppence__create_payment_link
  },
})) {
  console.log(message);
}

import { query } from "@anthropic-ai/claude-agent-sdk";
import { TUPPENCE_TOOLS, tuppenceMcpServer } from "@tuppence/agent-tools/claude-agent-sdk";

for await (const message of query({
  prompt: "Make a $15 payment link for a deep research report.",
  options: {
    mcpServers: { tuppence: tuppenceMcpServer() },
    allowedTools: [TUPPENCE_TOOLS], // or one by one: mcp__tuppence__create_payment_link
  },
})) {
  console.log(message);
}

import { query } from "@anthropic-ai/claude-agent-sdk";
import { TUPPENCE_TOOLS, tuppenceMcpServer } from "@tuppence/agent-tools/claude-agent-sdk";

for await (const message of query({
  prompt: "Make a €15 payment link for a deep research report.",
  options: {
    mcpServers: { tuppence: tuppenceMcpServer() },
    allowedTools: [TUPPENCE_TOOLS], // or one by one: mcp__tuppence__create_payment_link
  },
})) {
  console.log(message);
}

Each example runs in CI against the real API, with a scripted model in place of the LLM. The agent creates a real link and gets its URL back.

When the API says no

A refused call	The tool returns the API’s error (code, message, param) instead of crashing, so the agent can read it and correct the call.
Retries	Money-moving tools require an idempotency_key. A retry sends the same one, so the money moves once.

For one of your connected accounts

A platform’s agent can work for one of its connected accounts. Pass account when connecting and every tool acts as that account (its payments, links and bookings) using your key. The platform’s own Connect tools are then hidden.

Framework
OpenAI Agents SDK, LangChain	tuppenceTools({ account: "acct_…" })
Claude Agent SDK	tuppenceMcpServer({ account: "acct_…" })
@tuppence/mcp	TUPPENCE_ACCOUNT=acct_…

Without it, a single tool call can act as an account through its own account argument (MCP server).

Next

• MCP server: every tool, and how to connect Claude Desktop or Cursor.
• Agent identity: sign each request as the agent.