Where Is the Claude Code API Documentation?

Find the right official docs for Claude Code API questions: Anthropic API, Claude Code CLI, claude -p, Agent SDK, MCP, Managed Agents, auth, pricing, and automation.

If you are looking for “Claude Code API documentation,” first clarify which surface you mean. Claude Code is a CLI and coding agent workflow. Claude API is the platform API for calling models from your own application. The right documentation depends on whether you want model endpoints, terminal automation, a programmable agent, or external tool integrations.

Last checked on May 24, 2026. Claude Code and Claude Platform docs change quickly. Treat this FAQ as a routing guide, then verify exact endpoint names, flags, SDK syntax, auth rules, pricing, and limits in the official docs before shipping production code.

Short Answer

There is no single “Claude Code API” page that covers every developer use case.

If you mean...	Open this	Why
“I want to call Claude from my app.”	Anthropic API docs and API reference	This is where Messages API, models, streaming, tool use, rate limits, and SDKs live.
“I want to run Claude Code from a script.”	Claude Code CLI reference	This covers `claude`, `claude -p`, output formats, auth, update, and CLI flags.
“I want to build a coding agent in Python or TypeScript.”	Claude Agent SDK	This gives you Claude Code-style tools, sessions, MCP, hooks, permissions, and agent loops.
“I want Claude Code to access GitHub, docs, browser, databases, or internal systems.”	MCP docs	MCP is the tool/data-source integration layer.
“I want a hosted agent service.”	Claude Platform docs	Managed Agents are a hosted platform surface, not the local CLI.

The safest mental model is: Claude API for model calls, Claude Code CLI for terminal workflows, Agent SDK for programmable agents, MCP for external tools.

CLI vs Agent SDK vs Claude API

Surface	Best for	You manage
Claude Code CLI	Interactive local development, repo work, code edits, terminal workflows.	Repository state, approvals, local credentials, tool permissions.
`claude -p`	One-shot scripts, CI summaries, non-interactive checks.	Prompt scope, output parsing, auth, allowed tools, failure behavior.
Claude Agent SDK	Productized agents, custom automation, long-running sessions, programmatic approvals.	Hosting, sandboxing, keys, observability, tool policy, session lifecycle.
Anthropic API / client SDKs	Direct model calls in your own product.	Prompts, conversation state, tool loop, retries, rate limits, billing, UI.
Managed Agents	Hosted autonomous agents with Anthropic-managed infrastructure.	Agent definition, events, product integration, governance.

Do not mix these docs blindly. A flag from Claude Code CLI is not an HTTP endpoint. A Messages API request body is not an Agent SDK session config. An MCP server is not the same thing as the Claude model API.

Common Search Intent

“Claude Code API reference”

You probably need one of two things:

exact model API endpoints: open the Anthropic API reference;
Claude Code automation flags: open the Claude Code CLI reference.

For a broader routing page, use Claude Code API Documentation.

“Claude Code API docs”

If the task is about installing, logging in, running claude, using claude -p, adding MCP, or setting permissions, use Claude Code docs. If the task is about messages.create, model IDs, streaming, prompt caching, tool use, or API errors, use Claude Platform docs.

“Claude Code SDK”

Older pages may say “Claude Code SDK,” but the current official docs present this as the Claude Agent SDK. Use the Agent SDK when you want a Python or TypeScript program that can run Claude Code-style agents with file reading, command execution, editing, MCP, hooks, and permissions.

“Claude Code headless API”

For one-off non-interactive usage, start with:

claude -p "Summarize this pull request and list release risks"

For structured automation:

claude -p "Review this diff and return the top risks" \
  --output-format json \
  --permission-mode plan

For production-grade agents, use the Agent SDK rather than growing shell scripts forever.

Documentation Map

Task	Best starting page
Install Claude Code	Install Claude Code
Use CLI flags	Claude Code CLI reference
Run non-interactive prompts	Headless mode guide
Build an agent with SDK	Claude Agent SDK
Call Claude models from an app	Anthropic API docs
Find exact HTTP endpoints	Anthropic API reference
Add tool integrations	Claude Code MCP docs
Configure permissions	Tools allowlist
Control project behavior	CLAUDE.md guide
Estimate cost and limits	Claude Code pricing and limits

Authentication Rules

Authentication differs by surface:

Surface	Typical auth	Important note
Anthropic API	`ANTHROPIC_API_KEY` from Console	Best fit for production apps and backend services.
Claude Code CLI	Supported Claude Code login or Console/API billing flow	Good for local development and terminal workflows.
`claude -p`	Same CLI auth context unless configured otherwise	CI needs explicit credentials and permission boundaries.
Agent SDK	API key or supported provider credentials	Do not build customer-facing products around a personal interactive login.
MCP	Depends on each external server	Review scopes, secrets, and prompt injection risk before connecting.

For production automation, prefer explicit API-key or provider credentials, set spend limits, and keep secrets out of prompts, logs, CLAUDE.md, and repository files.

Common Mistakes

Mistake	Better approach
Treating ClaudeCode101 as an endpoint reference.	Use this site as a guide, then verify exact fields in official docs.
Looking for a single “Claude Code API” endpoint.	Decide whether you need API, CLI, SDK, or MCP.
Using `claude -p` for a complex product workflow.	Prototype with CLI, then move stable automation into Agent SDK.
Using Agent SDK for a one-line PR summary.	Use `claude -p` first.
Adding MCP when a simple API call or CLI command is enough.	Add MCP only when repeated external context is the bottleneck.
Running headless automation without budget limits.	Add narrow permissions, turn limits, logging, and spend controls.

Mini Decision Tree

Are you calling Claude from your own application? Use Anthropic API docs and client SDKs.
Are you asking Claude Code to work in a local repository? Use Claude Code CLI docs.
Do you need a script that runs once and exits? Use claude -p and the Headless Mode guide.
Do you need a programmatic agent with sessions, tools, hooks, and approvals? Use the Agent SDK.
Do you need Claude Code to access an external system repeatedly? Use MCP, but start read-only and review security first.