ToolContext from MCP Client is lost in Spring AI MCP Server

When using spring-ai-mcp (tested with 1.1.0-M3), the ToolContext values provided by the MCP client are not correctly propagated to the MCP server’s @Tool method.

This causes the ToolContext parameter injected into the tool method to lose the original client context, such as tokens or metadata.

---

Reproduction Steps

✅

MCP Client Code

chatClient.prompt()
    .toolContext(Map.of("token", request.getToken()))
    .user(request.getMessage())
    .stream()
    .content()
    .subscribe();

The client sets a token inside toolContext, expecting the MCP server to receive it.

❌

MCP Server Tool

@Tool(description = "Calculate financial statistics for a user")
public String calculateUserStatistics(Long createdByUserId, ToolContext toolContext) {
    // toolContext should contain "token" from client, but it’s missing
    System.out.println("Received ToolContext: " + toolContext);
    ...
}

On the server side, the toolContext received by the @Tool method does not contain the client-provided values (e.g. "token").

---

Root Cause Analysis

In the class

org.springframework.ai.mcp.McpToolUtils#toSharedSyncToolSpecification

the current code creates a new ToolContext as follows:

new ToolContext(Map.of(TOOL_CONTEXT_MCP_EXCHANGE_KEY, exchangeOrContext))

However, it does not include the meta field from the CallToolRequest (which contains the client’s tool context).

So the original client-provided ToolContext is completely ignored.

---

Expected Behavior

The ToolContext passed into the tool callback should merge both:

1. the existing MCP exchange context (TOOL_CONTEXT_MCP_EXCHANGE_KEY), and
2. the meta values from CallToolRequest (coming from the MCP client).

---

Proposed Fix

Modify McpToolUtils#toSharedSyncToolSpecification() to merge the client context:

Map<String, Object> mergedContext = new HashMap<>();
mergedContext.put(TOOL_CONTEXT_MCP_EXCHANGE_KEY, exchangeOrContext);
if (request.meta() != null) {
    mergedContext.putAll(request.meta());
}

ToolContext toolContext = new ToolContext(mergedContext);
String callResult = toolCallback.call(
    ModelOptionsUtils.toJsonString(request.arguments()), toolContext);

This will ensure that any context values set on the MCP client (via .toolContext(...)) are correctly available on the server.

---

Environment

• Spring AI version: 1.1.0-M3
• Java version: 25
• MCP client: Spring AI MCP Client
• MCP server: Spring AI MCP Server

---

Start in org.springframework.ai.mcp.McpToolUtils#toSharedSyncToolSpecification and inspect how CallToolRequest, request.meta(), and the exchange context are converted into ToolContext. Reproduce the MCP client/server flow with the token example, then verify that the @Tool method receives both the exchange context and client-provided metadata.