chore: set cache control markers on tools

intercept_anthropic_messages_base.go

@@ -1,9 +1,13 @@
 package aibridge
 
 import (
+	"fmt"
+	"net/http"
+	"net/http/httputil"
 	"strings"
 
 	"github.com/anthropics/anthropic-sdk-go"
+	"github.com/anthropics/anthropic-sdk-go/option"
 	"github.com/coder/aibridge/mcp"
 	"github.com/google/uuid"
 
@@ -44,6 +48,16 @@ func (i *AnthropicMessagesInterceptionBase) injectTools() {
 		return
 	}
 
+	// Any existing tool definitions.
+	for _, tool := range i.req.Tools {
+		if tool.OfTool == nil {
+			continue
+		}
+
+		// Explicitly unset all cache control settings, we'll set one at the end.
+		tool.OfTool.CacheControl = anthropic.CacheControlEphemeralParam{}
+	}
+
 	// Inject tools.
 	for _, tool := range i.mcpProxy.ListTools() {
 		i.req.Tools = append(i.req.Tools, anthropic.ToolUnionParam{
@@ -55,10 +69,18 @@ func (i *AnthropicMessagesInterceptionBase) injectTools() {
 				Name:        tool.ID,
 				Description: anthropic.String(tool.Description),
 				Type:        anthropic.ToolTypeCustom,
+				// Explicitly unset all cache control settings, we'll set one at the end.
+				CacheControl: anthropic.CacheControlEphemeralParam{},
 			},
 		})
 	}
 
+	// See https://docs.claude.com/en/docs/build-with-claude/prompt-caching.
+	// "The cache_control parameter on the last tool definition caches all tool definitions."
+	if count := len(i.req.Tools); count > 0 {
+		i.req.Tools[count-1].OfTool.CacheControl = anthropic.NewCacheControlEphemeralParam()
+	}
+
 	// Note: Parallel tool calls are disabled to avoid tool_use/tool_result block mismatches.
 	i.req.ToolChoice = anthropic.ToolChoiceUnionParam{
 		OfAny: &anthropic.ToolChoiceAnyParam{

intercept_anthropic_messages_streaming.go

@@ -61,6 +61,20 @@ func (i *AnthropicMessagesStreamingInterception) ProcessRequest(w http.ResponseW
 		return fmt.Errorf("developer error: req is nil")
 	}
 
+	// Explicitly unset any cache control markers on "assistant" messages; these should never be set
+	// since it's more beneficial for us to cache tool definitions, and Anthropic only allows for 4
+	// cache markers...
+	// https://docs.claude.com/en/docs/build-with-claude/prompt-caching#when-to-use-multiple-breakpoints
+	for _, msg := range i.req.Messages {
+		if msg.Role == anthropic.MessageParamRoleAssistant {
+			for _, c := range msg.Content {
+				if c.OfText != nil {
+					c.OfText.CacheControl = anthropic.CacheControlEphemeralParam{}
+				}
+			}
+		}
+	}
+
 	// Allow us to interrupt watch via cancel.
 	ctx, cancel := context.WithCancel(r.Context())
 	defer cancel()