From a1f2b999350acbf2ff6cbfacf8e4096c24944714 Mon Sep 17 00:00:00 2001
From: Lukasz Raczylo <lukasz@raczylo.com>
Date: Sun, 7 Jun 2026 13:32:32 +0100
Subject: [PATCH] test(e2e): solve_grid over real MCP stdio with fake LLM

---
 test/e2e/fakellm_test.go | 23 ++++++++++++
 test/e2e/main_test.go    | 18 ++++++++++
 test/e2e/server_test.go  | 78 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 119 insertions(+)
 create mode 100644 test/e2e/fakellm_test.go
 create mode 100644 test/e2e/main_test.go
 create mode 100644 test/e2e/server_test.go

diff --git a/test/e2e/fakellm_test.go b/test/e2e/fakellm_test.go
new file mode 100644
index 0000000..9196704
--- /dev/null
+++ b/test/e2e/fakellm_test.go
@@ -0,0 +1,23 @@
+//go:build e2e
+
+package e2e
+
+import (
+	"io"
+	"net/http"
+	"net/http/httptest"
+)
+
+// startFakeLLM returns an OpenAI-compatible endpoint that always selects tiles 0,1,2.
+func startFakeLLM() *httptest.Server {
+	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/v1/chat/completions":
+			_, _ = io.WriteString(w, `{"choices":[{"message":{"content":"{\"tiles\":[0,1,2],\"confidence\":0.95,\"recheck_recommended\":false}"}}]}`)
+		case "/v1/audio/transcriptions":
+			_, _ = io.WriteString(w, `{"text":"fortytwo"}`)
+		default:
+			http.NotFound(w, r)
+		}
+	}))
+}
diff --git a/test/e2e/main_test.go b/test/e2e/main_test.go
new file mode 100644
index 0000000..2e7877a
--- /dev/null
+++ b/test/e2e/main_test.go
@@ -0,0 +1,18 @@
+//go:build e2e
+
+package e2e
+
+import (
+	"os"
+	"os/exec"
+	"testing"
+)
+
+func TestMain(m *testing.M) {
+	build := exec.Command("go", "build", "-o", "../../bin/captcha-solver-mcp", "../../cmd/captcha-solver-mcp")
+	build.Stdout, build.Stderr = os.Stdout, os.Stderr
+	if err := build.Run(); err != nil {
+		os.Exit(1)
+	}
+	os.Exit(m.Run())
+}
diff --git a/test/e2e/server_test.go b/test/e2e/server_test.go
new file mode 100644
index 0000000..2a90964
--- /dev/null
+++ b/test/e2e/server_test.go
@@ -0,0 +1,78 @@
+//go:build e2e
+
+package e2e
+
+import (
+	"context"
+	"encoding/json"
+	"os/exec"
+	"testing"
+	"time"
+
+	"github.com/modelcontextprotocol/go-sdk/mcp"
+)
+
+func TestSolveGridOverStdio(t *testing.T) {
+	llmSrv := startFakeLLM()
+	defer llmSrv.Close()
+
+	cmd := exec.Command("../../bin/captcha-solver-mcp")
+	cmd.Env = append(cmd.Environ(),
+		"CAPTCHA_LLM_PROVIDER=openai-compatible",
+		"CAPTCHA_LLM_BASE_URL="+llmSrv.URL,
+		"CAPTCHA_LLM_API_KEY=test",
+		"CAPTCHA_LLM_MODEL=fake",
+	)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	defer cancel()
+
+	client := mcp.NewClient(&mcp.Implementation{Name: "e2e", Version: "v0"}, nil)
+	session, err := client.Connect(ctx, &mcp.CommandTransport{Command: cmd}, nil)
+	if err != nil {
+		t.Fatalf("connect: %v", err)
+	}
+	defer func() { _ = session.Close() }()
+
+	res, err := session.CallTool(ctx, &mcp.CallToolParams{
+		Name: "solve_grid_captcha",
+		Arguments: map[string]any{
+			"screenshot":   "iVBORw0KGgo=",
+			"instruction":  "select all buses",
+			"rows":         3,
+			"cols":         3,
+			"image_width":  300,
+			"image_height": 300,
+		},
+	})
+	if err != nil {
+		t.Fatalf("call: %v", err)
+	}
+	assertTiles(t, res, []int{0, 1, 2})
+}
+
+// assertTiles unmarshals the structured tool result and compares tiles.
+func assertTiles(t *testing.T, res *mcp.CallToolResult, want []int) {
+	t.Helper()
+	if res.IsError {
+		t.Fatalf("tool returned error: %+v", res.Content)
+	}
+	raw, err := json.Marshal(res.StructuredContent)
+	if err != nil {
+		t.Fatalf("marshal structured: %v", err)
+	}
+	var out struct {
+		Tiles []int `json:"tiles"`
+	}
+	if err := json.Unmarshal(raw, &out); err != nil {
+		t.Fatalf("unmarshal %q: %v", raw, err)
+	}
+	if len(out.Tiles) != len(want) {
+		t.Fatalf("tiles=%v want %v", out.Tiles, want)
+	}
+	for i := range want {
+		if out.Tiles[i] != want[i] {
+			t.Fatalf("tiles=%v want %v", out.Tiles, want)
+		}
+	}
+}