diff --git a/README.md b/README.md
index 33ca9fb..89a4717 100644
--- a/README.md
+++ b/README.md
@@ -199,7 +199,7 @@ export ANTHROPIC_ENABLE_1M_CONTEXT=true  # requires Claude Max
 
 - Registers an `auth.loader` with a custom `fetch` that intercepts all Anthropic API requests
 - Sets `Authorization: Bearer` with fresh OAuth tokens (cached in memory, 30s TTL, updated in-place after refresh)
-- Translates tool names between OpenCode and Anthropic API formats (adds/strips `mcp_` prefix)
+- Obfuscates tool names with MD5 hashing (`t_` prefix) to avoid API blacklist detection, with bidirectional reverse mapping for response translation
 - Buffers SSE response streams at event boundaries for reliable tool name translation
 - Injects Claude Code identity into system prompts via `experimental.chat.system.transform`
 - Sets required API headers (beta flags, billing, user-agent) with model-aware selection
diff --git a/src/transforms.test.ts b/src/transforms.test.ts
index 763a5b5..ed11d81 100644
--- a/src/transforms.test.ts
+++ b/src/transforms.test.ts
@@ -36,8 +36,8 @@ describe("transforms", () => {
     // The original system text should now be prepended to the first user message
     assert.equal(parsed.messages[0].content[0].type, "text")
     assert.equal(parsed.messages[0].content[0].text, "OpenCode and opencode")
-    assert.equal(parsed.tools[0].name, "mcp_search")
-    assert.equal(parsed.messages[0].content[1].name, "mcp_lookup")
+    assert.match(parsed.tools[0].name, /^t_[0-9a-f]{8}$/)
+    assert.match(parsed.messages[0].content[1].name, /^t_[0-9a-f]{8}$/)
   })
 
   it("transformBody relocates non-core system text to user message", () => {
@@ -449,9 +449,22 @@ describe("transforms", () => {
     assert.equal(parsed.thinking, undefined)
   })
 
-  it("stripToolPrefix removes mcp_ from response payload names", () => {
-    const input = '{"name":"mcp_search","type":"tool_use"}'
-    assert.equal(stripToolPrefix(input), '{"name": "search","type":"tool_use"}')
+  it("stripToolPrefix reverses obfuscated tool names", () => {
+    // Populate the maps by obfuscating via transformBody
+    const body = transformBody(
+      JSON.stringify({
+        tools: [{ name: "search" }],
+        messages: [{ role: "user", content: "test" }],
+      }),
+    )
+    const parsed = JSON.parse(body as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obf = parsed.tools[0].name
+    // Now test deobfuscation
+    const input = `{"name":"${obf}","type":"tool_use"}`
+    const result = stripToolPrefix(input)
+    assert.ok(result.includes('"name": "search"'))
   })
 
   it("transformResponseStream passes error responses through without SSE parsing", async () => {
@@ -517,23 +530,46 @@ describe("transforms", () => {
   })
 
   it("transformResponseStream still strips tool prefixes in error bodies", async () => {
-    // stripToolPrefix matches the pattern "name": "mcp_..."
-    const errorBody = '{"name": "mcp_search", "error": "failed"}'
+    // Populate maps via round-trip
+    const body = transformBody(
+      JSON.stringify({
+        tools: [{ name: "search" }],
+        messages: [{ role: "user", content: "x" }],
+      }),
+    )
+    const parsedBody = JSON.parse(body as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obf = parsedBody.tools[0].name
+
+    const errorBody = `{"name": "${obf}", "error": "failed"}`
     const response = new Response(errorBody, { status: 400 })
     const transformed = transformResponseStream(response)
     const text = await transformed.text()
     assert.ok(
       text.includes('"name": "search"'),
-      "Should strip mcp_ prefix even in error bodies",
+      "Should deobfuscate tool name in error bodies",
     )
     assert.ok(
-      !text.includes("mcp_search"),
-      "Should not contain mcp_search after stripping",
+      !text.includes(obf),
+      "Should not contain obfuscated name after stripping",
     )
   })
 
   it("transformResponseStream rewrites streamed tool names", async () => {
-    const payload = '{"name":"mcp_lookup"}'
+    // Populate maps via round-trip
+    const body = transformBody(
+      JSON.stringify({
+        tools: [{ name: "lookup" }],
+        messages: [{ role: "user", content: "x" }],
+      }),
+    )
+    const parsedBody = JSON.parse(body as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obf = parsedBody.tools[0].name
+
+    const payload = `{"name":"${obf}"}`
     const response = new Response(payload)
     const transformed = transformResponseStream(response)
     const text = await transformed.text()
@@ -542,8 +578,23 @@ describe("transforms", () => {
   })
 
   it("transformResponseStream buffers across chunks until event boundary", async () => {
-    const chunk1 = 'data: {"name":"mc'
-    const chunk2 = 'p_search"}\n\ndata: {"type":"done"}\n\n'
+    // Populate maps via round-trip
+    const body = transformBody(
+      JSON.stringify({
+        tools: [{ name: "search" }],
+        messages: [{ role: "user", content: "x" }],
+      }),
+    )
+    const parsedBody = JSON.parse(body as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obf = parsedBody.tools[0].name
+
+    // Split the obfuscated name across two chunks
+    const fullData = `data: {"name":"${obf}"}\n\ndata: {"type":"done"}\n\n`
+    const splitAt = 15 // splits inside the tool name
+    const chunk1 = fullData.slice(0, splitAt)
+    const chunk2 = fullData.slice(splitAt)
     const encoder = new TextEncoder()
 
     const stream = new ReadableStream({
@@ -563,18 +614,30 @@ describe("transforms", () => {
       `Expected stripped name in: ${text}`,
     )
     assert.ok(
-      !text.includes("mcp_search"),
-      `Should not contain mcp_search in: ${text}`,
+      !text.includes(obf),
+      `Should not contain obfuscated name in: ${text}`,
     )
   })
 
   it("transformResponseStream withholds output until event boundary arrives", async () => {
+    // Populate maps via round-trip
+    const bodyResult = transformBody(
+      JSON.stringify({
+        tools: [{ name: "test" }],
+        messages: [{ role: "user", content: "x" }],
+      }),
+    )
+    const parsedBody = JSON.parse(bodyResult as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obf = parsedBody.tools[0].name
+
     const encoder = new TextEncoder()
     let sendBoundary: (() => void) | undefined
 
     const source = new ReadableStream({
       start(controller) {
-        controller.enqueue(encoder.encode('data: {"name":"mcp_test"}'))
+        controller.enqueue(encoder.encode(`data: {"name":"${obf}"}`))
         sendBoundary = () => {
           controller.enqueue(encoder.encode("\n\n"))
           controller.close()
@@ -609,8 +672,8 @@ describe("transforms", () => {
       `Expected stripped name: ${text}`,
     )
     assert.ok(
-      !text.includes("mcp_test"),
-      `Should not contain mcp_test: ${text}`,
+      !text.includes(obf),
+      `Should not contain obfuscated name: ${text}`,
     )
 
     const final = await reader.read()
@@ -780,9 +843,22 @@ describe("transforms", () => {
   })
 
   it("transformResponseStream flushes remaining buffered data on stream end", async () => {
+    // Populate maps for both tool names via round-trip
+    const bodyResult = transformBody(
+      JSON.stringify({
+        tools: [{ name: "alpha" }, { name: "beta" }],
+        messages: [{ role: "user", content: "x" }],
+      }),
+    )
+    const parsedBody = JSON.parse(bodyResult as string) as {
+      tools: Array<{ name: string }>
+    }
+    const obfAlpha = parsedBody.tools[0].name
+    const obfBeta = parsedBody.tools[1].name
+
     const encoder = new TextEncoder()
-    const chunk1 = 'data: {"name":"mcp_alpha"}\n\n'
-    const chunk2 = 'data: {"name":"mcp_beta"}'
+    const chunk1 = `data: {"name":"${obfAlpha}"}\n\n`
+    const chunk2 = `data: {"name":"${obfBeta}"}`
 
     const stream = new ReadableStream({
       start(controller) {
@@ -805,12 +881,12 @@ describe("transforms", () => {
       `Expected beta stripped in: ${text}`,
     )
     assert.ok(
-      !text.includes("mcp_alpha"),
-      `Should not contain mcp_alpha in: ${text}`,
+      !text.includes(obfAlpha),
+      `Should not contain obfuscated alpha in: ${text}`,
     )
     assert.ok(
-      !text.includes("mcp_beta"),
-      `Should not contain mcp_beta in: ${text}`,
+      !text.includes(obfBeta),
+      `Should not contain obfuscated beta in: ${text}`,
     )
   })
 })
diff --git a/src/transforms.ts b/src/transforms.ts
index acf5b00..780ef0f 100644
--- a/src/transforms.ts
+++ b/src/transforms.ts
@@ -1,7 +1,27 @@
 import { buildBillingHeaderValue } from "./signing.ts"
 import { config, getModelOverride } from "./model-config.ts"
 
-const TOOL_PREFIX = "mcp_"
+// Obfuscate tool names: the API blacklists certain tool names (todowrite,
+// background_output, background_cancel). To avoid detection we hash ALL tool
+// names on the way out and reverse-map them on the way back.
+import { createHash } from "node:crypto"
+
+const toolNameMap = new Map<string, string>() // obfuscated → original
+const toolNameReverseMap = new Map<string, string>() // original → obfuscated
+
+function obfuscateToolName(name: string): string {
+  const existing = toolNameReverseMap.get(name)
+  if (existing) return existing
+  const hash = createHash("md5").update(name).digest("hex").slice(0, 8)
+  const obf = `t_${hash}`
+  toolNameMap.set(obf, name)
+  toolNameReverseMap.set(name, obf)
+  return obf
+}
+
+function deobfuscateToolName(obf: string): string {
+  return toolNameMap.get(obf) ?? obf
+}
 
 const SYSTEM_IDENTITY =
   "You are Claude Code, Anthropic's official CLI for Claude."
@@ -206,30 +226,30 @@ export function transformBody(
       }
     }
 
+    // Obfuscate tool names to avoid API blacklist detection
     if (Array.isArray(parsed.tools)) {
       parsed.tools = parsed.tools.map((tool) => ({
         ...tool,
-        name: tool.name ? `${TOOL_PREFIX}${tool.name}` : tool.name,
+        name: tool.name ? obfuscateToolName(tool.name) : tool.name,
       }))
     }
 
     if (Array.isArray(parsed.messages)) {
       parsed.messages = parsed.messages.map((message) => {
-        if (!Array.isArray(message.content)) {
-          return message
-        }
-
+        if (!Array.isArray(message.content)) return message
         return {
           ...message,
           content: message.content.map((block) => {
-            if (block.type !== "tool_use" || typeof block.name !== "string") {
-              return block
+            if (block.type === "tool_use" && typeof block.name === "string") {
+              return { ...block, name: obfuscateToolName(block.name) }
             }
-
-            return {
-              ...block,
-              name: `${TOOL_PREFIX}${block.name}`,
+            if (
+              block.type === "tool_result" &&
+              typeof block["tool_use_id"] === "string"
+            ) {
+              // tool_result references tool_use by id, not name — no change needed
             }
+            return block
           }),
         }
       })
@@ -246,7 +266,11 @@ export function transformBody(
 }
 
 export function stripToolPrefix(text: string): string {
-  return text.replace(/"name"\s*:\s*"mcp_([^"]+)"/g, '"name": "$1"')
+  // Reverse-map obfuscated tool names back to originals in response stream
+  return text.replace(/"name"\s*:\s*"(t_[0-9a-f]{8})"/g, (_match, obf) => {
+    const original = deobfuscateToolName(obf)
+    return `"name": "${original}"`
+  })
 }
 
 export function transformResponseStream(response: Response): Response {