--- title: "Anthropic (Claude)" description: "Create a bounded non-streaming Claude message, count input tokens and inspect accessible models using a customer Anthropic API key. Native content blocks, tool definitions and usage data are preserved. This four-command" --- > Documentation Index > Fetch the complete documentation index at: https://help.tooluse.net/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic (Claude)

Create a bounded non-streaming Claude message, count input tokens and inspect accessible models using a customer Anthropic API key. Native content blocks, tool definitions and usage data are preserved. This four-command REST subset excludes event streams, prompt-management beta APIs, files and asynchronous batches. Model availability and billing depend on the customer account.

What you can do (4)

What you can do

Available commands: 4 (2 read, 2 write). Enable only the commands your application needs.

Command Effect
Create one non-streaming message with a token ceiling — anthropic.messages.create Write
Count input tokens without generating an answer — anthropic.messages.count-tokens Write
List one page of accessible models — anthropic.models.list Read
Read one model description — anthropic.models.get Read

Connect and use

Connect and use

Open the integration's connection form to check the available credential source. The source has not been resolved for this documentation export. The instructions below apply when using your own provider account.

Get your provider credentials

Follow these steps for Anthropic (Claude), then enter the issued values in the connection form below.

  1. Open the Claude Console, select the intended workspace and create an API key in Settings → API keys. Configure workspace spend limits and only grant the required role. A Claude chat subscription alone does not provide API credits. Paste the key into this connection’s Anthropic (Claude) API token field. Start with models.list or messages.count-tokens; messages.create requires a model, messages and an explicit max_tokens ceiling. Streaming is disabled. Create a separate scoped Tooluse.net API key for your client; never expose the Anthropic key in the MCP URL.

Official credential setup

Where to enter credentials

In the integration page, sign in, choose your workspace and open Connect. If a Credentials choice is shown, select Use my credentials. Fill the following connection fields and save. Provider credentials belong in this form, never in an MCP configuration or tool input.

Connection field Requirement
Anthropic (Claude) API token Required

Open API keys and create a Tooluse.net key for the enabled commands.

Set TOOLUSE_API_KEY in your client and use the shared MCP configuration.

Configure Claude Code, Codex or Cursor once

Permissions and results

If you have more than one workspace, check the selected workspace in the connection or API-key dialog before saving. Roles, enabled commands, expiry and revocation use Tooluse.net's existing permissions. Disabling a connection or revoking a key immediately prevents new calls.

Send one Claude tool_use block per request. Keep the same Idempotency-Key when retrying the same request; use a new key for a new call. The response includes the shared Tooluse.net receipt and Claude tool_result. Check both HTTP status and the returned tool result before treating a call as successful.

Your connection's daily allowance and key permissions also apply. Adding a connection does not make a provider request.

Calls return the shared execution receipt and a Claude tool_result. Provider output varies by operation; consult the provider documentation for its response fields. Unconfirmed calls should be checked before retrying.

Call from your client

Use the shared MCP configuration. Refresh your client's tool list after saving the connection or changing key permissions, then call the returned tool name with the example input below.

Configure Claude Code, Codex or Cursor once

Discover your enabled tools

List permitted tools in Claude format. Read every page using page and size (maximum 100) until all results have been received. Use the returned name, not a catalog capability identifier, in the tool_use block.

curl 'https://api.tooluse.net/v1/tools' \
  -H "Authorization: Bearer $TOOLUSE_API_KEY"

Find this command in the tool list permitted by your key. Replace svc_FROM_YOUR_TOOL_LIST with its returned name. Use a new Idempotency-Key for each new call and retain it when retrying that same call.

HTTP call

curl 'https://api.tooluse.net/v1/tool-calls' \
  -H "Authorization: Bearer $TOOLUSE_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: replace-with-a-new-call-id' \
  --data '{
  "type": "tool_use",
  "id": "toolu_example",
  "name": "svc_FROM_YOUR_TOOL_LIST",
  "input": {}
}'

Send one Claude tool_use block per request. Keep the same Idempotency-Key when retrying the same request; use a new key for a new call. The response includes the shared Tooluse.net receipt and Claude tool_result. Check both HTTP status and the returned tool result before treating a call as successful.

Commands

Create one non-streaming message with a token ceiling — anthropic.messages.create

Run Create one non-streaming message with a token ceiling with Anthropic (Claude). This command can change provider data. Review the target and input before invoking it.

This operation can change provider data. Review the input before allowing an agent to call it.

Connect and use

Parameters

Requiredness is local to the containing object. A required child is required only when its parent object or array item is provided. Alternative schemas are identified in the Constraints column; the linked full JSON Schema is authoritative.

Parameter Type Required in parent Description Constraints
body object Required additionalProperties: false
body.model string Required minLength: 1
body.messages array Required minItems: 1
body.messages[] object Optional additionalProperties: false
body.messages[].role string Required enum: ["user","assistant"]
body.messages[].content — Required
body.messages[].content string Required oneOf[1]
body.messages[].content array Required oneOf[2]; minItems: 1
body.messages[].content[] object Optional oneOf[2]; additionalProperties: true
body.messages[].content[].type string Required oneOf[2]; minLength: 1
body.system — Optional
body.system string Optional oneOf[1]
body.system array Optional oneOf[2]
body.system[] object Optional oneOf[2]; additionalProperties: true
body.system[].type string Required oneOf[2]; minLength: 1
body.tools array Optional
body.tools[] object Optional additionalProperties: true
body.tools[].name string Required minLength: 1
body.tools[].description string Optional
body.tools[].input_schema object Required additionalProperties: true
body.tool_choice object Optional additionalProperties: false
body.tool_choice.type string Required enum: ["auto","any","tool","none"]
body.tool_choice.name string Optional
body.tool_choice.disable_parallel_tool_use boolean Optional
body.max_tokens integer Required minimum: 1; maximum: 64000
body.stream boolean Optional enum: [false]; default: false
body.temperature number Optional minimum: 0; maximum: 1
body.top_p number Optional minimum: 0; maximum: 1
body.top_k integer Optional minimum: 0
body.stop_sequences array Optional
body.stop_sequences[] string Optional
body.metadata object Optional additionalProperties: false
body.metadata.user_id string Optional maxLength: 256

Full JSON Schema

Full JSON Schema

Example input

{
  "body": {
    "model": "claude-sonnet-4-5",
    "max_tokens": 32,
    "messages": [
      {
        "role": "user",
        "content": "Say hello briefly."
      }
    ],
    "stream": false
  }
}

At most 1 provider requests per invocation; timeout: 30 seconds.

Count input tokens without generating an answer — anthropic.messages.count-tokens

Run Count input tokens without generating an answer with Anthropic (Claude). This command can change provider data. Review the target and input before invoking it.

This operation can change provider data. Review the input before allowing an agent to call it.

Connect and use

Parameters

Requiredness is local to the containing object. A required child is required only when its parent object or array item is provided. Alternative schemas are identified in the Constraints column; the linked full JSON Schema is authoritative.

Parameter Type Required in parent Description Constraints
body object Required additionalProperties: false
body.model string Required minLength: 1
body.messages array Required minItems: 1
body.messages[] object Optional additionalProperties: false
body.messages[].role string Required enum: ["user","assistant"]
body.messages[].content — Required
body.messages[].content string Required oneOf[1]
body.messages[].content array Required oneOf[2]; minItems: 1
body.messages[].content[] object Optional oneOf[2]; additionalProperties: true
body.messages[].content[].type string Required oneOf[2]; minLength: 1
body.system — Optional
body.system string Optional oneOf[1]
body.system array Optional oneOf[2]
body.system[] object Optional oneOf[2]; additionalProperties: true
body.system[].type string Required oneOf[2]; minLength: 1
body.tools array Optional
body.tools[] object Optional additionalProperties: true
body.tools[].name string Required minLength: 1
body.tools[].description string Optional
body.tools[].input_schema object Required additionalProperties: true
body.tool_choice object Optional additionalProperties: false
body.tool_choice.type string Required enum: ["auto","any","tool","none"]
body.tool_choice.name string Optional
body.tool_choice.disable_parallel_tool_use boolean Optional

Full JSON Schema

Full JSON Schema

Example input

{
  "body": {
    "model": "claude-sonnet-4-5",
    "messages": [
      {
        "role": "user",
        "content": "Count this input."
      }
    ]
  }
}

At most 1 provider requests per invocation; timeout: 30 seconds.

List one page of accessible models — anthropic.models.list

Run List one page of accessible models with Anthropic (Claude). This is a read command; it does not grant access beyond the connected account's permissions.

This is a read operation. Provider rate limits and data access rules still apply.

Connect and use

Parameters

Requiredness is local to the containing object. A required child is required only when its parent object or array item is provided. Alternative schemas are identified in the Constraints column; the linked full JSON Schema is authoritative.

Parameter Type Required in parent Description Constraints
query object Optional additionalProperties: false
query.before_id string Optional
query.after_id string Optional
query.limit integer Optional minimum: 1; maximum: 1000

Full JSON Schema

Full JSON Schema

Example input

{}

At most 1 provider requests per invocation; timeout: 30 seconds.

Read one model description — anthropic.models.get

Run Read one model description with Anthropic (Claude). This is a read command; it does not grant access beyond the connected account's permissions.

This is a read operation. Provider rate limits and data access rules still apply.

Connect and use

Parameters

Requiredness is local to the containing object. A required child is required only when its parent object or array item is provided. Alternative schemas are identified in the Constraints column; the linked full JSON Schema is authoritative.

Parameter Type Required in parent Description Constraints
path object Required additionalProperties: false
path.modelId string Required minLength: 1

Full JSON Schema

Full JSON Schema

Example input

{
  "path": {
    "modelId": "example"
  }
}

At most 1 provider requests per invocation; timeout: 30 seconds.

Errors

Check the key's expiry, revocation, workspace and allowed operations.

Reuse the same idempotency key only for the same request.

The connection or provider limit has been reached. Check the receipt before retrying.

Test the connection

Open the integration's commands below to inspect parameters and examples. Start with a read command for an account or resource you can access. Refresh your client's tool list, use the returned tool name, and check the execution result and usage. Provider charges and permissions still apply. Never test a write against production data without reviewing it.

Official provider documentation

Source: https://help.tooluse.net/integrations/anthropic/index.mdx