System Prompt

System Prompt

curl --request POST \
  --url https://api.anthropic.com/v1/messages/system-prompt \
  --header 'Content-Type: application/json' \
  --header 'anthropic-version: <anthropic-version>' \
  --header 'x-api-key: <api-key>' \
  --data '
{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "system": "You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.",
  "messages": [
    {
      "role": "user",
      "content": "What is a webhook?"
    }
  ]
}
'

import requests

url = "https://api.anthropic.com/v1/messages/system-prompt"

payload = {
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "system": "You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.",
    "messages": [
        {
            "role": "user",
            "content": "What is a webhook?"
        }
    ]
}
headers = {
    "anthropic-version": "<anthropic-version>",
    "x-api-key": "<api-key>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {
    'anthropic-version': '<anthropic-version>',
    'x-api-key': '<api-key>',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model: 'claude-sonnet-4-6',
    max_tokens: 1024,
    system: 'You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.',
    messages: [{role: 'user', content: 'What is a webhook?'}]
  })
};

fetch('https://api.anthropic.com/v1/messages/system-prompt', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

<?php

$curl = curl_init();

curl_setopt_array($curl, [
  CURLOPT_URL => "https://api.anthropic.com/v1/messages/system-prompt",
  CURLOPT_RETURNTRANSFER => true,
  CURLOPT_ENCODING => "",
  CURLOPT_MAXREDIRS => 10,
  CURLOPT_TIMEOUT => 30,
  CURLOPT_HTTP_VERSION => CURL_HTTP_VERSION_1_1,
  CURLOPT_CUSTOMREQUEST => "POST",
  CURLOPT_POSTFIELDS => json_encode([
    'model' => 'claude-sonnet-4-6',
    'max_tokens' => 1024,
    'system' => 'You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.',
    'messages' => [
        [
                'role' => 'user',
                'content' => 'What is a webhook?'
        ]
    ]
  ]),
  CURLOPT_HTTPHEADER => [
    "Content-Type: application/json",
    "anthropic-version: <anthropic-version>",
    "x-api-key: <api-key>"
  ],
]);

$response = curl_exec($curl);
$err = curl_error($curl);

curl_close($curl);

if ($err) {
  echo "cURL Error #:" . $err;
} else {
  echo $response;
}

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.anthropic.com/v1/messages/system-prompt"

	payload := strings.NewReader("{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("anthropic-version", "<anthropic-version>")
	req.Header.Add("x-api-key", "<api-key>")
	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

HttpResponse<String> response = Unirest.post("https://api.anthropic.com/v1/messages/system-prompt")
  .header("anthropic-version", "<anthropic-version>")
  .header("x-api-key", "<api-key>")
  .header("Content-Type", "application/json")
  .body("{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}")
  .asString();

require 'uri'
require 'net/http'

url = URI("https://api.anthropic.com/v1/messages/system-prompt")

http = Net::HTTP.new(url.host, url.port)
http.use_ssl = true

request = Net::HTTP::Post.new(url)
request["anthropic-version"] = '<anthropic-version>'
request["x-api-key"] = '<api-key>'
request["Content-Type"] = 'application/json'
request.body = "{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}"

response = http.request(request)
puts response.read_body

{
  "id": "msg_01XFDUDYJgAACzvnptvVoYEL",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "Hello! How can I help you today?"
    }
  ],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "stop_details": {},
  "usage": {
    "input_tokens": 17,
    "output_tokens": 12,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "service_tier": "standard",
    "inference_geo": "global"
  }
}

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "invalid x-api-key"
  },
  "request_id": "req_011CdDNcA4MfikfhDdfNYDNj"
}

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "invalid x-api-key"
  },
  "request_id": "req_011CdDNcA4MfikfhDdfNYDNj"
}

POST

messages

system-prompt

System Prompt

curl --request POST \
  --url https://api.anthropic.com/v1/messages/system-prompt \
  --header 'Content-Type: application/json' \
  --header 'anthropic-version: <anthropic-version>' \
  --header 'x-api-key: <api-key>' \
  --data '
{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "system": "You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.",
  "messages": [
    {
      "role": "user",
      "content": "What is a webhook?"
    }
  ]
}
'

import requests

url = "https://api.anthropic.com/v1/messages/system-prompt"

payload = {
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "system": "You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.",
    "messages": [
        {
            "role": "user",
            "content": "What is a webhook?"
        }
    ]
}
headers = {
    "anthropic-version": "<anthropic-version>",
    "x-api-key": "<api-key>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {
    'anthropic-version': '<anthropic-version>',
    'x-api-key': '<api-key>',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model: 'claude-sonnet-4-6',
    max_tokens: 1024,
    system: 'You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.',
    messages: [{role: 'user', content: 'What is a webhook?'}]
  })
};

fetch('https://api.anthropic.com/v1/messages/system-prompt', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

<?php

$curl = curl_init();

curl_setopt_array($curl, [
  CURLOPT_URL => "https://api.anthropic.com/v1/messages/system-prompt",
  CURLOPT_RETURNTRANSFER => true,
  CURLOPT_ENCODING => "",
  CURLOPT_MAXREDIRS => 10,
  CURLOPT_TIMEOUT => 30,
  CURLOPT_HTTP_VERSION => CURL_HTTP_VERSION_1_1,
  CURLOPT_CUSTOMREQUEST => "POST",
  CURLOPT_POSTFIELDS => json_encode([
    'model' => 'claude-sonnet-4-6',
    'max_tokens' => 1024,
    'system' => 'You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.',
    'messages' => [
        [
                'role' => 'user',
                'content' => 'What is a webhook?'
        ]
    ]
  ]),
  CURLOPT_HTTPHEADER => [
    "Content-Type: application/json",
    "anthropic-version: <anthropic-version>",
    "x-api-key: <api-key>"
  ],
]);

$response = curl_exec($curl);
$err = curl_error($curl);

curl_close($curl);

if ($err) {
  echo "cURL Error #:" . $err;
} else {
  echo $response;
}

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.anthropic.com/v1/messages/system-prompt"

	payload := strings.NewReader("{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("anthropic-version", "<anthropic-version>")
	req.Header.Add("x-api-key", "<api-key>")
	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

HttpResponse<String> response = Unirest.post("https://api.anthropic.com/v1/messages/system-prompt")
  .header("anthropic-version", "<anthropic-version>")
  .header("x-api-key", "<api-key>")
  .header("Content-Type", "application/json")
  .body("{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}")
  .asString();

require 'uri'
require 'net/http'

url = URI("https://api.anthropic.com/v1/messages/system-prompt")

http = Net::HTTP.new(url.host, url.port)
http.use_ssl = true

request = Net::HTTP::Post.new(url)
request["anthropic-version"] = '<anthropic-version>'
request["x-api-key"] = '<api-key>'
request["Content-Type"] = 'application/json'
request.body = "{\n  \"model\": \"claude-sonnet-4-6\",\n  \"max_tokens\": 1024,\n  \"system\": \"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences.\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": \"What is a webhook?\"\n    }\n  ]\n}"

response = http.request(request)
puts response.read_body

{
  "id": "msg_01XFDUDYJgAACzvnptvVoYEL",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "Hello! How can I help you today?"
    }
  ],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "stop_details": {},
  "usage": {
    "input_tokens": 17,
    "output_tokens": 12,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "service_tier": "standard",
    "inference_geo": "global"
  }
}

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "invalid x-api-key"
  },
  "request_id": "req_011CdDNcA4MfikfhDdfNYDNj"
}

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "invalid x-api-key"
  },
  "request_id": "req_011CdDNcA4MfikfhDdfNYDNj"
}

Overview

A system prompt is a top-level instruction you give Claude before the conversation begins. It defines Claude’s role, tone, scope, and constraints for the entire interaction. Unlike user messages, the system prompt sits outside the messages array — it is passed as a separate system field in the request body.

How the System Prompt Works

The system prompt is processed before any user message. Claude reads it first and uses it as the lens through which it interprets everything that follows.

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "system": "You are a fintech API documentation assistant. Explain financial technology concepts clearly for developers. Keep responses under three sentences.",
  "messages": [
    {
      "role": "user",
      "content": "What is a webhook?"
    }
  ]
}

What the System Prompt Controls

Behavior	Example Instruction
Role definition	`"You are a senior DevOps engineer..."`
Tone and style	`"Always respond formally and concisely..."`
Scope limiting	`"Only answer questions about our API..."`
Output format	`"Always respond in valid JSON..."`
Length control	`"Keep all responses under three sentences."`
Safety boundaries	`"Never discuss competitor products..."`

Token Impact

The system prompt counts as input tokens on every request — because it is included in full each time. A verbose system prompt adds cost at scale.

Request	Input Tokens	System Prompt
No system prompt	17	None
With system prompt	53	284 characters

Write system prompts that are specific and concise. Every unnecessary word costs tokens on every single request.

What the System Prompt Cannot Do

System prompts shape behavior but do not override Claude’s core guidelines. Claude will decline requests that violate Anthropic’s usage policies regardless of what the system prompt instructs.

A system prompt that instructs Claude to produce harmful content, impersonate real individuals, or bypass its guidelines will not work as intended. Claude’s values are not configurable.

Real Test Results

These two responses were produced from the same user message — "What is a webhook?" — with different system prompts: Fintech assistant system prompt:

“A webhook is a way for one system to automatically notify another system when a specific event happens, by sending an HTTP POST request to a URL you provide.”

No system prompt:

“Hello! How can I help you today?”

The same question. Entirely different responses. The system prompt is doing all the work.

Token counts confirm this — the fintech system prompt request used 53 input tokens vs 17 for the basic message. The 36 token difference is the system prompt itself being processed.

Messages

Understand how system prompts relate to the messages array

Tokens

How system prompt length affects input token costs

Multi-turn

How system prompts persist across multi-turn conversations

Tool Use

Combine system prompts with tools for focused AI agents

Authorizations

x-api-key

string

header

required

Headers

anthropic-version

string

required

Example:

"2023-06-01"

Body

application/json

model

enum<string>

required

The Claude model to use.

Available options:

claude-sonnet-4-6,

claude-opus-4-6,

claude-haiku-4-5

Example:

"claude-sonnet-4-6"

max_tokens

integer

required

Maximum number of output tokens to generate.

Required range: x >= 1

Example:

1024

messages

object[]

required

Show child attributes

system

string

System-level instructions that shape Claude's behavior throughout the conversation.

Example:

"You are a fintech API documentation assistant. You explain financial technology concepts clearly and concisely for developers. Always respond in plain English, avoid jargon, and keep responses under three sentences."

Response

Successful response

string

Example:

"msg_01XFDUDYJgAACzvnptvVoYEL"

type

string

Example:

"message"

role

string

Example:

"assistant"

content

object[]

Show child attributes

model

string

Example:

"claude-sonnet-4-6"

stop_reason

enum<string>

Available options:

end_turn,

max_tokens,

stop_sequence,

tool_use

Example:

"end_turn"

stop_details

object | null

usage

object

Show child attributes

Send a Message Multi-turn Conversation

Endpoints

Overview

How the System Prompt Works

What the System Prompt Controls

Token Impact

What the System Prompt Cannot Do

Real Test Results

Further Reading

Messages

Tokens

Multi-turn

Tool Use

Authorizations

Headers

Body

Response

​Overview

​How the System Prompt Works

​What the System Prompt Controls

​Token Impact

​What the System Prompt Cannot Do

​Real Test Results

​Further Reading

Messages

Tokens

Multi-turn

Tool Use

Authorizations

Headers

Body

Response

Overview

How the System Prompt Works

What the System Prompt Controls

Token Impact

What the System Prompt Cannot Do

Real Test Results

Further Reading