MCP, the Model Context Protocol, is the open standard that lets AI assistants use external services through tools. Everything API to Agents does is described with the 19 terms below, each defined once and linked to the page that goes deeper.

Model Context Protocol (MCP)

An open protocol that lets AI assistants and agents connect to external tools, data, and services through a standard interface. An assistant that supports MCP can use any MCP server without custom integration code for each service.

MCP server

A program that exposes tools (and optionally resources and prompts) over the Model Context Protocol so that MCP clients can call them. For a SaaS company, an MCP server is the layer between the product API and the assistants customers use.

How an API becomes an MCP server

Remote MCP server

An MCP server reached over HTTPS at a URL, rather than run locally on the user's machine. ChatGPT, Claude, and other hosted assistants connect to remote MCP servers; a managed MCP server is always remote.

MCP client

The assistant-side software that connects to MCP servers and calls their tools on the user's behalf. ChatGPT, Claude, Claude Code, and various IDE agents are MCP clients.

MCP tool

A single capability an MCP server publishes: a name, a description that tells the assistant when to use it, a typed input schema, and a result. Tools are the unit an assistant reasons about, so a good integration exposes a few outcome-shaped tools rather than one per API endpoint.

Writing tool descriptions

Read tool

A tool that retrieves information without changing anything: search a catalogue, look up an order, fetch a document. Read tools still need permission checks so one customer never sees another's records.

Read-only tools still need permissions

Write tool

A tool that creates or changes data: add an item, update a record, create a request. Write tools need an idempotency contract so a retried call does not duplicate the action.

Idempotency for agent actions

Guarded tool

A write tool whose action has real consequences, such as placing an order, cancelling a booking, or changing billing. The assistant must obtain an explicit user confirmation before the server executes it.

Tool description

The natural-language text attached to a tool that tells an assistant what the tool does, when to use it, and when not to. It is the main lever for an assistant choosing the right tool.

Agent readiness

How prepared an API is to be used safely by an AI agent: documentation quality, authentication and safety, a clear tool and workflow model, and a way to test. API to Agents scores it out of 100 across those four dimensions.

The free Agent Readiness Audit

Agent Readiness Audit

API to Agents' free, automated assessment. It reads API documentation, returns a readiness score, proposed MCP tools, risks with mitigations, a roadmap with a delivery estimate, and a fixed setup quote, in about two minutes and without credentials.

What the audit delivers

Managed MCP server

An MCP server that a provider designs, hosts, secures, monitors, and maintains for a customer's API, served from a subdomain the customer controls. API to Agents offers it for a fixed setup fee and a monthly hosting plan.

Managed MCP server

Managed MCP hosting

The recurring part of a managed MCP server: running the endpoint, monitoring it, and updating it as MCP clients change. API to Agents plans start at $29.99 per month and are billed only once the integration is live.

Hosting plans

Setup quote

The one-time price to design, build, test, and launch an MCP server for an API. At API to Agents it is calculated by versioned pricing rules from the audit's proposed scope, starts at $499, and is valid for 14 days.

Every line item

Upstream API / upstream system

The customer's existing service that the MCP server calls: a REST or GraphQL API, a webhook-driven system, an SDK, or a private integration API. A second upstream system is a priced scope item.

OpenAPI to MCP

Converting an OpenAPI description into MCP tool definitions. Operation names, parameters, schemas, and security schemes convert mechanically; tool selection, descriptions, result shaping, and safety boundaries still need design.

What can be automated

Connector

The term ChatGPT and Claude use for a configured connection to a remote MCP server. A user adds the server's URL in the assistant's settings and authenticates; the assistant then calls the server's tools.

Claude integration

Continuation token (pagination)

An opaque value a tool returns when a result set is larger than one page, so the assistant can ask for the next page without inventing offsets and can tell the user when it has only seen part of the data.

MCP tool pagination

Evaluation set

A repeatable collection of realistic user goals, fixtures, allowed actions, and expected outcomes used to test an MCP integration before and after changes.

Building an evaluation set