How to Connect Anthropic Claude API to Make.com: Complete Scenario Setup

Why Integrate Claude into Make.com via Native HTTP Modules?

Make.com’s app store features several community-built Claude integrations. However, relying on community apps for enterprise production pipelines presents three critical bottlenecks:

1. Zero Day-One Availability: When Anthropic updates its models (such as rolling out upgraded Sonnet or Haiku checkpoints), third-party wrappers can take weeks or months to expose the new model strings in their dropdowns. By using Make’s native HTTP: Make a request module, you can switch models in under 5 seconds by simply updating a JSON string.
2. Full Parameter Control: Community modules often hide advanced Anthropic parameters like stop_sequences, top_k, top_p, and vision image payloads. A raw HTTP request gives you 100% specification parity with the official Anthropic Messages API.
3. Sovereign Credential Security: Passing API tokens through unverified community apps introduces unnecessary compliance risks. Direct HTTP calls connect your Make.com execution worker directly to api.anthropic.com over TLS 1.3 encryption.

Before configuring the endpoint, explore our curated directory of Best Workflow Automation Software and evaluate model performance benchmarks in our showdowns on ChatGPT vs Claude and Claude 3.5 Sonnet vs GPT-4o for Coding.

Step 1: Provisioning Your Anthropic API Key

To authenticate requests, you require a production secret key:

1. Visit the Anthropic Developer Console and sign in.
2. Navigate to Settings → API Keys.
3. Click Create Key, name it Make-Production-Worker, and copy the secret token immediately (it begins with sk-ant-api03-).
4. Navigate to Plans & Billing and add at least \$10 in prepaid credits. Anthropic does not operate on a free tier for automated API calls; unbilled accounts will throw an immediate 400 Credit Balance Too Low error.

Step 2: Configuring the Make.com HTTP “Make a request” Module

Open your scenario in Make.com and add the HTTP module with the Make a request action:

Field	Production Value	Technical Explanation
URL	https://api.anthropic.com/v1/messages	Official Messages API endpoint
Method	POST	Standard REST creation method
Headers (1)	x-api-key: sk-ant-api03-...	Anthropic secret bearer token
Headers (2)	anthropic-version: 2023-06-01	Mandatory API versioning header
Headers (3)	content-type: application/json	Declares JSON request format
Body Type	Raw	Enables manual JSON construction
Content Type	JSON (application/json)	Serializes outgoing payload
Parse response	Yes	Automatically converts response to Make data bundle

Step 3: Constructing the JSON Payload with System Prompts

Unlike legacy completion APIs, Anthropic enforces a strict separation between system prompts and the messages array. Enter the following JSON template into the Request content box:

{
  "model": "claude-3-5-sonnet-20241022",
  "max_tokens": 2048,
  "temperature": 0.2,
  "system": "You are an elite data extraction assistant. Analyze the incoming customer email and extract structured entities into strict JSON. Never include markdown code fences or conversational greetings. Output only valid JSON.",
  "messages": [
    {
      "role": "user",
      "content": "Extract the lead data from this message: " + 1.email_body
    }
  ]
}

Choosing Between Sonnet, Haiku, and Opus in Make.com

• Claude 3.5 Sonnet (claude-3-5-sonnet-20241022): The gold standard for complex workflows requiring logical reasoning, code execution, multi-step math, and nuanced decision-making (\$3.00 / million input tokens).
• Claude 3.5 Haiku (claude-3-5-haiku-20241022): Blazing-fast inference speeds with near-instant response times. Ideal for high-volume categorization, sentiment tagging, and basic field extraction at a fraction of the cost (\$0.80 / million input tokens).
• Claude 3 Opus (claude-3-opus-20240229): High-capacity model for deeply nuanced academic or long-form prose synthesis (\$15.00 / million input tokens).

Step 4: Forcing 100% Reliable JSON Output (Assistant Prefill Technique)

One of the most powerful features of the Anthropic API is Assistant Message Prefilling. While models occasionally output conversational commentary like “Sure! Here is the JSON you requested:”, you can completely eliminate this behavior by prefilling the assistant’s turn with an opening curly brace {:

{
  "model": "claude-3-5-sonnet-20241022",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Classify this customer ticket: " + 1.ticket_text
    },
    {
      "role": "assistant",
      "content": "{"
    }
  ]
}

Because the assistant’s response begins with {, Claude is programmatically constrained to continue writing the valid JSON object body, completely eliminating markdown fences (```json) and conversational chatter.

Step 5: Parsing and Downstream Field Mapping

When the HTTP module completes, Claude’s generated answer resides at:
{2.data.content[1].text} (or prepended with { if using the prefill trick).

1. Add a JSON: Parse JSON module immediately following the HTTP step.
2. In the JSON string field, supply: "{" + 2.data.content[1].text (if using prefill) or 2.data.content[1].text.
3. Click Run this module only with a sample payload to establish the data schema.
4. Make.com will inspect the JSON keys (e.g., lead_score, urgency, recommended_action) and expose them as native selectable pills in subsequent modules (like Slack, Airtable, or HubSpot).

Handling Production Rate Limits and Edge Cases

As your automation scales, Anthropic will enforce concurrency and token rate limits (RPM and TPM). Protect your scenario against crashes by implementing resilient error handling:

• Attach a Break Directive on HTTP 429: If Anthropic responds with a rate limit, attach an Error Handler ending in Break. This automatically saves the bundle to your Incomplete Executions queue and retries with exponential backoff after 1, 5, and 15 minutes. See our definitive guide to Make.com Error Handlers Explained.
• Handling 500 & 529 Overloaded Errors: During major global AI events, Anthropic’s clusters may experience brief capacity constraints (HTTP 529). Combine the Break directive with a 2-minute delay to ensure no customer transactions are lost.
• Token Usage Budgeting: For agencies running high-volume automations, track operational margins using our interactive SaaS Burn Rate Calculator.

Claude 3.5 Sonnet vs OpenAI GPT-4o: Cost & Latency Benchmark in Make.com

Model	Input / Output Cost (1M Tokens)	Context Window	Average Latency	Make.com Best Fit
Claude 3.5 Sonnet	$3.00 / $15.00	200,000 tokens	1.4s	Complex logic, code generation, multi-step agent flows
Claude 3.5 Haiku	$0.80 / $4.00	200,000 tokens	0.4s	High-volume triage, JSON parsing, sentiment tagging
GPT-4o	$2.50 / $10.00	128,000 tokens	1.1s	General multi-modal extraction, voice synthesis
GPT-4o mini	$0.15 / $0.60	128,000 tokens	0.6s	Ultra-budget classifications

Multi-Modal Vision: Ingesting PDFs and Images with Claude in Make.com

Claude 3.5 Sonnet can analyze binary images, charts, and diagrams directly inside your Make.com workflow. To submit an image:

1. Pass your image through a Base64 Encode function inside Make.com: {toBase64(1.data)}.
2. Construct the multi-modal content block inside your user message payload:

{
  "model": "claude-3-5-sonnet-20241022",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "image",
          "source": {
            "type": "base64",
            "media_type": "image/jpeg",
            "data": "BASE64_STRING_HERE"
          }
        },
        {
          "type": "text",
          "text": "Extract all numbers from this dashboard screenshot."
        }
      ]
    }
  ]
}

This unlocks automated receipt auditing, ad creative inspection, and document verification without needing external OCR software. For an alternative open-source implementation, see our guide on Make vs n8n.

Advanced Pattern: Function Calling (Tool Use) in Make.com

Anthropic’s Messages API supports declarative Tool Use, allowing Claude to intelligently decide when to invoke external APIs, look up customer records, or perform calculations.

To implement tool calling in your Make.com scenario:

1. Define your tools array inside the raw JSON payload passed to the HTTP module:

   {
     "model": "claude-3-5-sonnet-20241022",
     "max_tokens": 1024,
     "tools": [
       {
         "name": "lookup_customer_crm",
         "description": "Searches the CRM database for customer subscription status by email.",
         "input_schema": {
           "type": "object",
           "properties": {
             "email": {
               "type": "string",
               "description": "The customer's email address"
             }
           },
           "required": ["email"]
         }
       }
     ],
     "messages": [
       {
         "role": "user",
         "content": "Check the status of billing for customer alex@example.com."
       }
     ]
   }

2. When Claude determines a tool should be executed, it responds with stop_reason: "tool_use" and an invocation block containing id, name, and input parameters.
3. In Make.com, attach a Router:
   • Path A (Standard Response): Filter on stop_reason = "end_turn". Return generated text to user.
   • Path B (Tool Execution): Filter on stop_reason = "tool_use". Route to your HubSpot / Airtable lookup module using {2.data.content[1].input.email}.
   • Path C (Tool Result Return): Execute a secondary HTTP call back to Anthropic passing role: "user" with type: "tool_result" containing the data fetched from your database.

Prompt Caching: Slashing Latency and Costs by up to 90%

If your Make.com scenario repeatedly analyzes lengthy brand guidelines, large codebase files, or dense corporate knowledge bases, using Anthropic’s Prompt Caching dramatically cuts operational costs:

• Input Token Discount: Cached tokens receive a 90% discount (\$0.30 per million tokens on Sonnet vs \$3.00 normal).
• Latency Reduction: Time-to-first-token drops by up to 80% because Anthropic does not re-tokenize the static system prompt.
• Configuration in Make.com: Add the anthropic-beta: prompt-caching-2024-07-31 header to your HTTP module, and append "cache_control": {"type": "ephemeral"} to your system prompt block. The cache persists for 5 minutes and automatically refreshes whenever called.

Synchronous vs Streaming Responses in Make.com Workflows

Anthropic supports server-sent events (SSE) streaming for real-time token delivery. However, Make.com is fundamentally an atomic, transaction-based workflow engine.

When configuring your HTTP request:

• Keep "stream": false (Default): Never set "stream": true inside Make.com scenarios. Streaming emits partial chunk events that Make.com cannot assemble natively without complex custom webhook listeners. Leaving stream disabled returns the complete completed message in a single atomic payload.
• Timeout Thresholds: Claude 3.5 Sonnet processing large context windows (50,000+ words) may take 15 to 25 seconds to generate deep analytical responses. Ensure your Make.com HTTP module’s “Timeout” setting is increased from the default 30 seconds to 60 or 120 seconds to prevent premature client-side execution timeouts.

Diagnostic Matrix: Common Anthropic API Errors in Make.com

HTTP Status	Error Type	Root Cause	Immediate Solution
400 Bad Request	invalid_request_error	Missing anthropic-version header or malformed JSON	Add header anthropic-version: 2023-06-01 and validate JSON syntax
401 Unauthorized	authentication_error	Invalid or revoked x-api-key	Regenerate key in console.anthropic.com and re-enter in Make.com
429 Rate Limit	rate_limit_error	Exceeded RPM (Requests per min) or TPM (Tokens per min)	Attach error handler with Break directive for exponential retry
529 Overloaded	overloaded_error	Anthropic server clusters experiencing high global traffic	Queue executions with Break directive or fallback to Claude 3.5 Haiku