tessl/maven-dev-langchain4j--langchain4j-anthropic

This package provides an integration layer between the LangChain4j framework and Anthropic's Claude language models, enabling Java developers to seamlessly incorporate Anthropic's AI capabilities into their applications.

Overview

Eval results

Files

Tool Support

Name: tessl/maven-dev-langchain4j--langchain4j-anthropic
Author: tessl

The langchain4j-anthropic integration supports both user-defined tools (functions the model can call) and Anthropic server-side tools (like web search executed by Anthropic's servers).

User-Defined Tools

User-defined tools are configured using ToolSpecification from langchain4j-core. These tools are executed by your application code.

Basic Tool Usage

import dev.langchain4j.agent.tool.ToolSpecification;
import dev.langchain4j.agent.tool.ToolExecutionRequest;
import dev.langchain4j.data.message.ToolExecutionResultMessage;
import dev.langchain4j.model.chat.request.json.JsonObjectSchema;

// Define a tool
ToolSpecification weatherTool = ToolSpecification.builder()
    .name("get_weather")
    .description("Get current weather for a location")
    .parameters(JsonObjectSchema.builder()
        .addStringProperty("location", "City name")
        .required("location")
        .build())
    .build();

// Create model with tool
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .toolSpecifications(weatherTool)
    .build();

// Chat and handle tool calls
ChatResponse response = model.chat(ChatRequest.builder()
    .messages(UserMessage.from("What's the weather in Paris?"))
    .build());

if (response.aiMessage().hasToolExecutionRequests()) {
    for (ToolExecutionRequest request : response.aiMessage().toolExecutionRequests()) {
        String toolName = request.name();
        String arguments = request.arguments();

        // Execute tool
        String result = executeWeatherTool(toolName, arguments);

        // Send result back
        ChatResponse finalResponse = model.chat(ChatRequest.builder()
            .messages(
                UserMessage.from("What's the weather in Paris?"),
                response.aiMessage(),
                ToolExecutionResultMessage.from(request, result)
            )
            .build());

        System.out.println(finalResponse.aiMessage().text());
    }
}

Error Handling:

RuntimeException: Tool name not found in registered tools
RuntimeException: Tool arguments don't match schema (if strictTools enabled)
RuntimeException: Tool execution fails in user code (handle in application)

Common Pitfalls:

❌ DON'T forget to send tool results back

if (response.aiMessage().hasToolExecutionRequests()) {
    executeTools(response.aiMessage().toolExecutionRequests());
    // Missing: send results back to model
}

✅ DO complete the tool loop

if (response.aiMessage().hasToolExecutionRequests()) {
    List<ToolExecutionRequest> requests = response.aiMessage().toolExecutionRequests();
    String result = executeTools(requests);
    ChatResponse nextResponse = model.chat(ChatRequest.builder()
        .messages(userMsg, aiMsg, ToolExecutionResultMessage.from(requests.get(0), result))
        .build());
}

Performance Notes:

Each tool call requires additional API round-trip
Parallel tool use reduces latency for multi-tool scenarios
Tool schemas increase request size; keep descriptions concise

Tool Choice Strategies

Control which tools the model uses.

package dev.langchain4j.model.chat.request;

/**
 * Strategy for tool selection.
 */
public enum ToolChoice {
    /** Model decides whether to use tools - default behavior */
    AUTO,

    /** Model must use at least one tool from the available set */
    REQUIRED,

    /** Model cannot use any tools */
    NONE;
}

Usage Example:

import dev.langchain4j.model.chat.request.ToolChoice;

// Auto: Model decides whether to use tools
AnthropicChatModel autoModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .toolChoice(ToolChoice.AUTO)
    .build();

// Required: Model must use at least one tool
AnthropicChatModel requiredAnyModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .toolChoice(ToolChoice.REQUIRED)  // Requires use of at least one tool
    .build();

// Required: Model must use a specific tool
AnthropicChatModel requiredModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(weatherTool)
    .toolChoice(ToolChoice.REQUIRED)
    .toolChoiceName("get_weather")
    .build();

// None: Model cannot use tools
AnthropicChatModel noneModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .toolChoice(ToolChoice.NONE)
    .build();

Error Handling:

IllegalArgumentException: toolChoiceName set without REQUIRED toolChoice
RuntimeException: Specified tool name not found in toolSpecifications
RuntimeException: REQUIRED set but no tools will match user query

Common Pitfalls:

❌ DON'T use REQUIRED without appropriate tools

.toolSpecifications(calculatorTool)
.toolChoice(ToolChoice.REQUIRED)
// User asks: "What's the weather?" - model forced to use calculator

✅ DO match tools to expected use cases

.toolSpecifications(weatherTool, timeToolget, newsToolget)
.toolChoice(ToolChoice.AUTO)  // Model chooses appropriate tool

Use Cases by Strategy:

AUTO: General purpose, model decides when tools help
REQUIRED: Structured workflows where tool use is mandatory
NONE: Disable tools temporarily while keeping model configuration

Parallel vs Sequential Tool Execution

Control whether the model can call multiple tools in parallel.

// Enable parallel tool execution (default)
AnthropicChatModel parallelModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .disableParallelToolUse(false)
    .build();

// Disable parallel tool execution
AnthropicChatModel sequentialModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .disableParallelToolUse(true)
    .build();

Error Handling:

No specific errors; configuration applied to all requests
Parallel tool calls may fail individually (handle in tool execution code)

Common Pitfalls:

❌ DON'T disable parallel when tools are independent

.disableParallelToolUse(true)  // Wastes time on independent tool calls

✅ DO enable parallel for independent tools

.disableParallelToolUse(false)  // Weather and time can be fetched simultaneously

Performance Comparison:

Parallel enabled: Multiple tool calls in single response (lower latency)
Parallel disabled: One tool call per response (sequential rounds)
Latency reduction: 50-80% for multi-tool scenarios with parallel enabled

When to Disable Parallel:

Tools have dependencies (output of one needed by another)
Rate limiting concerns (e.g., external API quotas)
Stateful tools where order matters

Strict Tool Schema Validation

Enable strict validation of tool schemas.

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tools)
    .strictTools(true)  // Enable strict schema validation
    .build();

Error Handling:

RuntimeException: Tool arguments fail schema validation
RuntimeException: Missing required parameters
RuntimeException: Type mismatch in parameters

Common Pitfalls:

❌ DON'T enable strict without thorough schema testing

.strictTools(true)  // May reject valid tool calls due to schema issues

✅ DO test schemas thoroughly first

// Test with strictTools(false), then enable
.strictTools(false)  // Development
.strictTools(true)   // Production after validation

Validation Behavior:

Strict disabled: API attempts best-effort coercion (e.g., "123" → 123)
Strict enabled: API rejects type mismatches, fails fast
Trade-off: Strict catches errors earlier but less forgiving

Tool Metadata

Pass additional metadata with tool specifications. Tool metadata allows you to send custom information about tools to the model, which can help the model make better decisions about when and how to use tools.

The toolMetadataKeysToSend builder method controls which metadata keys from the ToolSpecification metadata map are sent to the Anthropic API. By default, no metadata is sent. You can specify which keys to include.

Common Use Cases:

Categorizing tools by domain or function type
Including version information for tool compatibility
Adding priority or cost hints for tool selection
Providing context about tool reliability or performance

package dev.langchain4j.agent.tool;

import java.util.Map;

/**
 * Tool specification with metadata support.
 */
public class ToolSpecification {
    /**
     * Returns tool metadata map.
     *
     * @return metadata map, never null (may be empty)
     */
    public Map<String, Object> metadata();
}

Usage Example:

import java.util.Set;

ToolSpecification tool = ToolSpecification.builder()
    .name("calculator")
    .description("Perform calculations")
    .parameters(schema)
    .metadata(Map.of(
        "category", "math",
        "version", "1.0",
        "cost", "low",
        "priority", "high"
    ))
    .build();

// Send specific metadata keys
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tool)
    .toolMetadataKeysToSend(Set.of("category"))  // Only send "category"
    .build();

// Or use varargs to send multiple keys
AnthropicChatModel model2 = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .toolSpecifications(tool)
    .toolMetadataKeysToSend("category", "version", "priority")
    .build();

Error Handling:

No errors if specified keys not present in metadata (silently ignored)
Large metadata increases request size (monitor costs)

Common Pitfalls:

❌ DON'T send all metadata keys

.toolMetadataKeysToSend(tool.metadata().keySet())  // May include internal keys

✅ DO explicitly control which keys to send

.toolMetadataKeysToSend("category", "version")  // Only relevant keys

Performance Notes:

Each metadata key increases request size (~10-50 bytes per key)
More metadata may improve tool selection quality
Balance between helpfulness and request size

Using with AI Services

Tools integrate seamlessly with LangChain4j's AI Services.

import dev.langchain4j.service.AiServices;
import dev.langchain4j.agent.tool.Tool;

class Calculator {
    @Tool("Add two numbers")
    int add(int a, int b) {
        return a + b;
    }

    @Tool("Multiply two numbers")
    int multiply(int a, int b) {
        return a * b;
    }
}

interface MathAssistant {
    String chat(String message);
}

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .build();

MathAssistant assistant = AiServices.builder(MathAssistant.class)
    .chatLanguageModel(model)
    .tools(new Calculator())
    .build();

String response = assistant.chat("What is 15 + 27 multiplied by 3?");
System.out.println(response);

AI Services Benefits:

Automatic tool discovery via @Tool annotations
Tool execution handled automatically
Conversation history managed transparently
Exception handling and retries built-in

Error Handling:

Tool execution exceptions propagated to caller
Automatic retry on transient failures
Configurable timeout per tool

Server Tools

Server tools are executed by Anthropic's servers (e.g., web search, code execution). These are experimental features.

AnthropicServerTool

Define server-side tools that Anthropic will execute.

package dev.langchain4j.model.anthropic;

import java.util.Map;

/**
 * Server-side tool executed by Anthropic's infrastructure.
 * EXPERIMENTAL: API may change.
 *
 * @since 1.0.0
 */
public class AnthropicServerTool {
    /**
     * Creates new builder for server tool.
     *
     * @return new builder instance, never null
     */
    public static Builder builder();

    /**
     * Returns tool type identifier with version.
     *
     * @return type string (e.g., "web_search_20250305"), never null
     */
    public String type();

    /**
     * Returns tool name for identification.
     *
     * @return tool name, never null
     */
    public String name();

    /**
     * Returns tool configuration attributes.
     *
     * @return attributes map, never null (may be empty)
     */
    public Map<String, Object> attributes();

    /**
     * Compares server tools for equality.
     *
     * @param o other object
     * @return true if equal
     */
    public boolean equals(Object o);

    /**
     * Returns hash code.
     *
     * @return hash code
     */
    public int hashCode();

    /**
     * Returns string representation.
     *
     * @return string form
     */
    public String toString();

    /**
     * Builder for AnthropicServerTool.
     */
    public static class Builder {
        /**
         * Sets tool type with version suffix.
         *
         * @param type tool type identifier, must not be null
         * @return this builder, never null
         * @throws IllegalArgumentException if type is null
         * @default no default - REQUIRED
         */
        public Builder type(String type);

        /**
         * Sets tool name.
         *
         * @param name tool name, must not be null
         * @return this builder, never null
         * @throws IllegalArgumentException if name is null
         * @default no default - REQUIRED
         */
        public Builder name(String name);

        /**
         * Sets all attributes at once.
         *
         * @param attributes attribute map, may be null
         * @return this builder, never null
         * @default empty map
         */
        public Builder attributes(Map<String, Object> attributes);

        /**
         * Adds single attribute.
         *
         * @param key attribute name, must not be null
         * @param value attribute value, may be null
         * @return this builder, never null
         * @throws IllegalArgumentException if key is null
         * @default no default
         */
        public Builder addAttribute(String key, Object value);

        /**
         * Builds server tool instance.
         *
         * @return configured server tool, never null
         * @throws IllegalStateException if type or name missing
         */
        public AnthropicServerTool build();
    }
}

Web Search Tool

Configure web search as a server tool.

import dev.langchain4j.model.anthropic.AnthropicServerTool;
import java.util.List;

AnthropicServerTool webSearch = AnthropicServerTool.builder()
    .type("web_search_20250305")
    .name("web_search")
    .addAttribute("max_uses", 5)
    .addAttribute("allowed_domains", List.of(
        "wikipedia.org",
        "britannica.com"
    ))
    .build();

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .serverTools(webSearch)
    .returnServerToolResults(true)  // Include results in response
    .build();

ChatResponse response = model.chat(ChatRequest.builder()
    .messages(UserMessage.from("What is the population of Tokyo?"))
    .build());

// Access server tool results
Map<String, Object> attributes = response.aiMessage().attributes();
if (attributes.containsKey("server_tool_results")) {
    @SuppressWarnings("unchecked")
    List<AnthropicServerToolResult> results =
        (List<AnthropicServerToolResult>) attributes.get("server_tool_results");

    for (AnthropicServerToolResult result : results) {
        System.out.println("Tool: " + result.type());
        System.out.println("Content: " + result.content());
    }
}

Web Search Attributes:

max_uses (Integer): Maximum searches per request (1-10 recommended)
allowed_domains (List<String>): Whitelist of domains to search

Error Handling:

RuntimeException: Web search not available for API key
RuntimeException: max_uses exceeded
RuntimeException: No results found (non-fatal, empty results returned)

Common Pitfalls:

❌ DON'T set max_uses too high

.addAttribute("max_uses", 100)  // Expensive, slow, often unnecessary

✅ DO use reasonable limits

.addAttribute("max_uses", 3)  // Usually sufficient

Performance Notes:

Each search adds 2-5 seconds latency
Results cached by Anthropic (within session)
Cost per search ~equivalent to 1000-2000 tokens

Code Execution Tool

Configure the code execution server tool for running code snippets.

import dev.langchain4j.model.anthropic.AnthropicServerTool;

AnthropicServerTool codeExecution = AnthropicServerTool.builder()
    .type("code_execution_20250305")
    .name("code_execution")
    .addAttribute("max_uses", 5)  // Maximum number of executions per request
    .build();

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .serverTools(codeExecution)
    .returnServerToolResults(true)  // Include execution output
    .build();

ChatResponse response = model.chat(ChatRequest.builder()
    .messages(UserMessage.from("Calculate the first 10 Fibonacci numbers"))
    .build());

Code Execution Attributes:

max_uses (Integer): Maximum number of code executions allowed in a single request

Code Execution Results: Code execution results include:

stdout: Standard output from the executed code
stderr: Standard error output
return_code: Exit code (0 for success)

Error Handling:

RuntimeException: Code execution not available for API key
RuntimeException: max_uses exceeded
Non-zero return_code in results (not an exception, check in application)

Security Notes:

Code runs in sandboxed environment (Python 3)
No network access
Limited execution time (~30 seconds max)
No file system persistence

Common Pitfalls:

❌ DON'T assume unlimited execution time

// Long-running code may timeout
"while True: calculate()"  // Will be terminated

✅ DO keep executions short

// Fast, focused computations
"result = sum(range(100))"  // Completes quickly

Multiple Server Tools

Configure multiple server tools simultaneously.

AnthropicServerTool webSearch = AnthropicServerTool.builder()
    .type("web_search_20250305")
    .name("web_search")
    .addAttribute("max_uses", 3)
    .addAttribute("allowed_domains", List.of("wikipedia.org", "github.com"))
    .build();

AnthropicServerTool codeExecution = AnthropicServerTool.builder()
    .type("code_execution_20250305")
    .name("code_execution")
    .addAttribute("max_uses", 2)
    .build();

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(modelName)
    .serverTools(webSearch, codeExecution)
    .returnServerToolResults(true)
    .build();

ChatResponse response = model.chat(ChatRequest.builder()
    .messages(UserMessage.from("Search for Python sorting algorithms and show me an implementation"))
    .build());

Error Handling:

Individual tool failures don't fail entire request
Check each result's type and content for errors

Performance Notes:

Multiple server tools increase max latency significantly
Total max_uses applies across all tools (manage budgets carefully)

AnthropicServerToolResult

Results from server-executed tools.

package dev.langchain4j.model.anthropic;

/**
 * Result from server-side tool execution.
 *
 * @since 1.0.0
 */
public class AnthropicServerToolResult {
    /**
     * Creates new builder.
     *
     * @return new builder, never null
     */
    public static Builder builder();

    /**
     * Returns result type identifier.
     *
     * @return type string (e.g., "web_search_tool_result"), never null
     */
    public String type();

    /**
     * Returns tool execution ID linking to original request.
     *
     * @return tool use ID, never null
     */
    public String toolUseId();

    /**
     * Returns result content (type varies by tool).
     * Web search: List<Map<String, Object>> with title, url, snippet.
     * Code execution: Map<String, Object> with stdout, stderr, return_code.
     *
     * @return result content, never null
     */
    public Object content();

    /**
     * Compares results for equality.
     *
     * @param o other object
     * @return true if equal
     */
    public boolean equals(Object o);

    /**
     * Returns hash code.
     *
     * @return hash code
     */
    public int hashCode();

    /**
     * Returns string representation.
     *
     * @return string form
     */
    public String toString();

    /**
     * Builder for server tool result.
     */
    public static class Builder {
        /**
         * Sets result type.
         *
         * @param type result type identifier, must not be null
         * @return this builder, never null
         * @default no default - REQUIRED
         */
        public Builder type(String type);

        /**
         * Sets tool use ID.
         *
         * @param toolUseId execution ID, must not be null
         * @return this builder, never null
         * @default no default - REQUIRED
         */
        public Builder toolUseId(String toolUseId);

        /**
         * Sets result content.
         *
         * @param content result data, must not be null
         * @return this builder, never null
         * @default no default - REQUIRED
         */
        public Builder content(Object content);

        /**
         * Builds result instance.
         *
         * @return configured result, never null
         * @throws IllegalStateException if required fields missing
         */
        public AnthropicServerToolResult build();
    }
}

Processing Server Tool Results

Handle different types of server tool results.

@SuppressWarnings("unchecked")
List<AnthropicServerToolResult> results =
    (List<AnthropicServerToolResult>) response.aiMessage()
        .attributes()
        .get("server_tool_results");

for (AnthropicServerToolResult result : results) {
    String type = result.type();

    if (type.equals("web_search_tool_result")) {
        // Web search results are typically List<Map<String, Object>>
        List<Map<String, Object>> searchResults =
            (List<Map<String, Object>>) result.content();

        for (Map<String, Object> searchResult : searchResults) {
            System.out.println("Title: " + searchResult.get("title"));
            System.out.println("URL: " + searchResult.get("url"));
            System.out.println("Snippet: " + searchResult.get("snippet"));
        }
    } else if (type.equals("code_execution_tool_result")) {
        // Code execution results are typically Map<String, Object>
        Map<String, Object> execResult =
            (Map<String, Object>) result.content();

        System.out.println("Stdout: " + execResult.get("stdout"));
        System.out.println("Stderr: " + execResult.get("stderr"));
        System.out.println("Return code: " + execResult.get("return_code"));
    }
}

Error Handling:

Check stderr for code execution errors
Check return_code (0 = success, non-zero = error)
Empty search results not an error (just no matches found)

Null Safety:

content() never null but may contain null values
Cast safely with @SuppressWarnings("unchecked") after type check
Validate expected fields present before accessing

Types

ToolSpecification (from langchain4j-core)

package dev.langchain4j.agent.tool;

import dev.langchain4j.model.chat.request.json.JsonSchema;
import java.util.Map;

/**
 * Specification for a user-defined tool.
 */
public class ToolSpecification {
    /**
     * Creates new builder.
     *
     * @return new builder, never null
     */
    public static Builder builder();

    /**
     * Returns tool name.
     *
     * @return name, never null
     */
    public String name();

    /**
     * Returns tool description.
     *
     * @return description, never null
     */
    public String description();

    /**
     * Returns parameter schema.
     *
     * @return JSON schema, may be null
     */
    public JsonSchema parameters();

    /**
     * Returns tool metadata.
     *
     * @return metadata map, never null (may be empty)
     */
    public Map<String, Object> metadata();
}

ToolExecutionRequest (from langchain4j-core)

package dev.langchain4j.data.message;

/**
 * Request to execute a tool from the model.
 */
public class ToolExecutionRequest {
    /**
     * Returns unique execution request ID.
     *
     * @return request ID, never null
     */
    public String id();

    /**
     * Returns tool name to execute.
     *
     * @return tool name, never null
     */
    public String name();

    /**
     * Returns JSON-encoded tool arguments.
     *
     * @return arguments JSON string, never null
     */
    public String arguments();
}

ToolChoice

package dev.langchain4j.model.chat.request;

/**
 * Strategy for tool selection.
 */
public enum ToolChoice {
    /** Model decides whether to use tools */
    AUTO,

    /** Model must use at least one tool */
    REQUIRED,

    /** Model cannot use tools */
    NONE;
}

Server Tool Types Reference

Web Search (web_search_20250305)

Attributes:

max_uses (Integer): Maximum number of searches per request (validation: 1-10)
allowed_domains (List<String>): Restrict searches to specific domains (validation: max 20 domains)

Result Format:

List<Map<String, Object>> where each map contains:
- "title" (String): Page title, never null
- "url" (String): Page URL, never null
- "snippet" (String): Text excerpt, never null

Null Safety:

All result fields are never null
Empty list if no results found

Code Execution (code_execution_20250305)

Attributes:

max_uses (Integer): Maximum number of code executions per request (validation: >= 1)

Result Format:

Map<String, Object> containing:
- "stdout" (String): Standard output, never null (may be empty)
- "stderr" (String): Error output, never null (may be empty)
- "return_code" (Integer): Exit code (0 = success), never null

Null Safety:

All result fields are never null
Check return_code to determine success

Notes

User-defined tools are executed by your application code
Server tools are executed by Anthropic's servers (e.g., web_search, code_execution)
Server tools are experimental and may change
Tool results can be large; returnServerToolResults is disabled by default
Web search results typically include title, URL, and snippet
Code execution runs in a sandboxed environment
Code execution results include stdout, stderr, and return code
Parallel tool execution allows the model to call multiple tools simultaneously
Strict tools enforce schema validation at API level
Tool metadata provides additional context to help the model select appropriate tools
toolMetadataKeysToSend controls which metadata keys are sent to the API
Tools work identically in streaming mode
Server tool types use version suffixes (e.g., _20250305) for API versioning

Thread Safety:

ToolSpecification is immutable and thread-safe
AnthropicServerTool is immutable and thread-safe
Tool execution code must handle concurrency if tools reused across threads

Install with Tessl CLI

npx tessl i tessl/maven-dev-langchain4j--langchain4j-anthropic@1.11.0

docs

streaming-chat-model.md

token-count-estimator.md

tools.md

tile.json

tessl/maven-dev-langchain4j--langchain4j-anthropic

tools.md.css-3qkkll{font-size:var(--chakra-font-sizes-sm);font-weight:var(--chakra-font-weights-normal);color:var(--chakra-colors-gray-300);}docs/

Tool Support

User-Defined Tools

Basic Tool Usage

Tool Choice Strategies

Parallel vs Sequential Tool Execution

Strict Tool Schema Validation

Tool Metadata

Using with AI Services

Server Tools

AnthropicServerTool

Web Search Tool

Code Execution Tool

Multiple Server Tools

AnthropicServerToolResult

Processing Server Tool Results

Types

ToolSpecification (from langchain4j-core)

ToolExecutionRequest (from langchain4j-core)

ToolChoice

Server Tool Types Reference

Web Search (web_search_20250305)

Code Execution (code_execution_20250305)

Notes

tools.mddocs/