fix(cursor): HTTP/2 AgentService support + version bump to 3.12.17

Real Cursor IDE now uses AgentService at agent.api5.cursor.sh (HTTP/2-only)
while 9router still spoke the retired ChatService at api2.cursor.sh with
outdated headers, producing HTTP 429 "Update Required". Add an executeAgent
path that builds an agent.v1.RunRequest Connect RPC over a raw http2 stream
and fetches the account-specific usable model catalog via GetUsableModels.

Also implement MCP tool calling over AgentService: encode OpenAI tools as
AgentRunRequest.mcp_tools (McpToolDefinition with google.protobuf.Value
input_schema), decode McpArgs tool calls, and forward them to the client as
OpenAI tool_calls so the client runs the tool and resumes in the next turn.
Reply to request_context_args with a non-empty RequestContext, to server
heartbeats with client_heartbeat, and to KV blob get/set with empty results,
so action queries no longer stall the stream. Fold the client system prompt
into the user message (custom_system_prompt makes the server return an empty
turn). Bump clientVersion to 3.12.17 and add the x-cursor-client-commit
header so the gateway identifies as a current Cursor IDE release.

tests/unit/cursor-models.test.js

@@ -0,0 +1,103 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import {
+  clearCursorModelCache,
+  parseCursorUsableModels,
+  resolveCursorModels,
+} from "../../open-sse/services/cursorModels.js";
+
+const originalFetch = global.fetch;
+
+function varint(value) {
+  const bytes = [];
+  while (value >= 0x80) {
+    bytes.push((value & 0x7f) | 0x80);
+    value >>>= 7;
+  }
+  bytes.push(value);
+  return Uint8Array.from(bytes);
+}
+
+function field(fieldNumber, value) {
+  return Uint8Array.from([(fieldNumber << 3) | 2, ...varint(value.length), ...value]);
+}
+
+function text(value) {
+  return new TextEncoder().encode(value);
+}
+
+function concat(...parts) {
+  const size = parts.reduce((sum, part) => sum + part.length, 0);
+  const result = new Uint8Array(size);
+  let offset = 0;
+  for (const part of parts) {
+    result.set(part, offset);
+    offset += part.length;
+  }
+  return result;
+}
+
+function model(id, name) {
+  return field(1, concat(field(1, text(id)), field(4, text(name))));
+}
+
+describe("Cursor live model catalog", () => {
+  beforeEach(() => {
+    clearCursorModelCache();
+  });
+
+  afterEach(() => {
+    global.fetch = originalFetch;
+    clearCursorModelCache();
+  });
+
+  it("decodes the GetUsableModels protobuf response", () => {
+    const payload = concat(
+      model("default", "Auto"),
+      model("gpt-5.3-codex", "GPT 5.3 Codex"),
+      model("gpt-5.3-codex", "Duplicate"),
+    );
+
+    expect(parseCursorUsableModels(payload)).toEqual([
+      { id: "default", name: "Auto" },
+      { id: "gpt-5.3-codex", name: "GPT 5.3 Codex" },
+    ]);
+  });
+
+  it("fetches the account-specific catalog and caches it", async () => {
+    const payload = concat(model("claude-4.6-opus", "Claude 4.6 Opus"));
+    global.fetch = vi.fn().mockResolvedValue(new Response(payload, { status: 200 }));
+    const credentials = {
+      accessToken: "cursor-token",
+      providerSpecificData: { machineId: "machine-id" },
+    };
+
+    await expect(resolveCursorModels(credentials)).resolves.toEqual({
+      models: [{ id: "claude-4.6-opus", name: "Claude 4.6 Opus" }],
+    });
+    await expect(resolveCursorModels(credentials)).resolves.toEqual({
+      models: [{ id: "claude-4.6-opus", name: "Claude 4.6 Opus" }],
+    });
+
+    expect(global.fetch).toHaveBeenCalledTimes(1);
+    expect(global.fetch).toHaveBeenCalledWith(
+      "https://agent.api5.cursor.sh/agent.v1.AgentService/GetUsableModels",
+      expect.objectContaining({
+        method: "POST",
+        body: expect.any(Uint8Array),
+        headers: expect.objectContaining({
+          "content-type": "application/proto",
+          accept: "application/proto",
+        }),
+      }),
+    );
+  });
+
+  it("fails open when the Cursor catalog request fails", async () => {
+    global.fetch = vi.fn().mockResolvedValue(new Response("no", { status: 403 }));
+
+    await expect(resolveCursorModels({
+      accessToken: "cursor-token",
+      providerSpecificData: { machineId: "machine-id" },
+    })).resolves.toBeNull();
+  });
+});