Skip to main content

MCP Tools

Ottu Core ships a built-in MCP (Model Context Protocol) server — a single connection that exposes a curated set of tools to AI agents and MCP-compatible clients (Claude, Cursor, Google ADK, and others). Instead of stitching REST calls together by hand, an agent connects once, discovers the available tools, and calls them by name. Every tool is permission-gated and returns clean, JSON-native results.

The first tool documented here powers Agentic merchant onboarding. Before an agent can configure a merchant's payment gateways, it needs to know which gateways and banks are actually available in the merchant's country. The PG Capabilities tool answers exactly that — it lists every payment gateway (PG) per country, the banks and providers behind each one, and what each gateway can do.

What is MCP?

The Model Context Protocol is an open standard that lets an AI application discover and call tools over one connection. Ottu implements the server side; your agent framework provides the client. You describe the goal in natural language and the client decides which tool to call.

Connecting

The Ottu Core MCP server is served over Streamable HTTP at the /mcp/ path of your Ottu instance:

https://sandbox.ottu.net/mcp/

Swap in your own merchant domain when you integrate. Point any Streamable-HTTP MCP client at that URL and it will negotiate the session and list the available tools for you.

Authentication

Every request must be authenticated as an Ottu user with a Bearer token:

Authorization: Bearer <your-token>

The token identifies the user making the call, and each tool runs with that user's permissions — the same permissions that govern the equivalent action in the dashboard. For how to obtain and manage credentials, see Authentication.

Permissions

Each tool enforces its own permission at the boundary. A caller that lacks the required permission gets a clean permission-denied result rather than partial data or a misleading error. For example, PG Capabilities requires the gateway.view_pgmid permission.

Error responses

Tools never surface raw exceptions to the agent. When something is wrong — a missing permission, an unknown argument, or a not-found lookup — the tool returns a small JSON object with a single error key and a human-readable message:

Error result
{ "error": "You do not have permission to view gateway capabilities." }

A successful call returns the tool's structured result, documented per tool. Your MCP client surfaces this as the tool's return value.

Available tools

ToolPurposePermission
get_pg_capabilitiesList payment gateways and the banks/providers that support them, per country, with each gateway's funding sources, integration types, capabilities, supported wallets, and SSL requirement.gateway.view_pgmid

What's Next?