SSH MCP Server

import { MCPSSHServer } from "../src/mcp-ssh-server.js"; import { SSHConnectionManager } from "../src/ssh-connection-manager.js"; describe("MCP Command Interception", () => { let mcpServer: MCPSSHServer; let sshManager: SSHConnectionManager; const testSessionName = "interception-test-session"; beforeEach(async () => { sshManager = new SSHConnectionManager(); mcpServer = new MCPSSHServer({}, sshManager); // Create real SSH connection using CLAUDE.md compliant approach (no mocks) const connectionResult = await mcpServer.callTool("ssh_connect", { name: testSessionName, host: "localhost", username: "jsbattig", keyFilePath: "/home/jsbattig/.ssh/id_ed25519" }); // Verify connection was successful if (!(connectionResult as any).success) { throw new Error(`Failed to establish SSH connection: ${(connectionResult as any).error}`); } }); afterEach(async () => { // Cleanup real SSH connection try { await mcpServer.callTool("ssh_disconnect", { sessionName: testSessionName }); } catch (error) { console.warn(`Warning: Error during SSH cleanup: ${error}`); } }); describe("Buffer Content Detection", () => { it("should intercept MCP ssh_exec when browser commands are buffered", async () => { // Arrange: Add commands to browser buffer sshManager.addBrowserCommand(testSessionName, "pwd", "cmd-1", "user"); sshManager.addBrowserCommand(testSessionName, "whoami", "cmd-2", "user"); // Act: Try to execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "ls -la" }); // Assert: Should get CommandGatingError with complete command objects expect(result).toEqual({ success: false, error: "BROWSER_COMMANDS_EXECUTED", message: "User executed commands directly in browser", browserCommands: [ { command: "pwd", commandId: "cmd-1", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 // -1 indicates command not yet completed } }, { command: "whoami", commandId: "cmd-2", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 // -1 indicates command not yet completed } } ], retryAllowed: true }); }); it("should raise CommandGatingError when browser command buffer contains content", async () => { // Arrange: Add single command to browser buffer sshManager.addBrowserCommand(testSessionName, "echo 'test'", "cmd-1", "user"); // Act: Try to execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "date" }); // Assert: Should get structured error response expect(result).toEqual({ success: false, error: "BROWSER_COMMANDS_EXECUTED", message: "User executed commands directly in browser", browserCommands: [ { command: "echo 'test'", commandId: "cmd-1", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 // -1 indicates command not yet completed } } ], retryAllowed: true }); }); }); describe("Empty Buffer Pass-Through", () => { it("should proceed normally when browser command buffer is empty", async () => { // Arrange: Ensure buffer is empty sshManager.clearBrowserCommandBuffer(testSessionName); // Act: Execute command via MCP (real SSH execution) const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "echo 'test'" }); // Assert: Should get standard MCP success response with real command output expect(result).toEqual({ success: true, result: { stdout: expect.stringContaining("test"), stderr: "", exitCode: 0 } }); }); it("should return standard MCP response when buffer is empty", async () => { // Arrange: Explicitly clear buffer sshManager.clearBrowserCommandBuffer(testSessionName); // Act: Execute command via MCP (real SSH execution) const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "echo 'Hello World'" }); // Assert: Standard success response format with real command output expect(result).toEqual({ success: true, result: { stdout: expect.stringContaining("Hello World"), stderr: "", exitCode: 0 } }); }); }); describe("Multiple Buffered Commands", () => { it("should include complete command array in error response", async () => { // Arrange: Add multiple commands to buffer sshManager.addBrowserCommand(testSessionName, "pwd", "cmd-1", "user"); sshManager.addBrowserCommand(testSessionName, "ls -la", "cmd-2", "user"); sshManager.addBrowserCommand(testSessionName, "whoami", "cmd-3", "user"); // Act: Try to execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "date" }); // Assert: Should include all buffered commands expect(result).toEqual({ success: false, error: "BROWSER_COMMANDS_EXECUTED", message: "User executed commands directly in browser", browserCommands: [ { command: "pwd", commandId: "cmd-1", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } }, { command: "ls -la", commandId: "cmd-2", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } }, { command: "whoami", commandId: "cmd-3", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } } ], retryAllowed: true }); }); it("should clear buffer after interception error response", async () => { // Arrange: Add commands to buffer const commands = ["git status", "git log --oneline", "git branch"]; commands.forEach((cmd, i) => { sshManager.addBrowserCommand(testSessionName, cmd, `cmd-${i}`, "user"); }); // Verify buffer has content before interception expect(sshManager.getBrowserCommandBuffer(testSessionName)).toHaveLength(3); // Act: Try to execute command (should be intercepted) const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "echo 'blocked'" }); // Assert: Should get error response and buffer cleared expect(result).toMatchObject({ success: false, error: "BROWSER_COMMANDS_EXECUTED" }); // Buffer should be cleared after error response const bufferAfterInterception = sshManager.getBrowserCommandBuffer(testSessionName); expect(bufferAfterInterception).toHaveLength(0); }); it("should handle mixed user and claude commands in buffer", async () => { // Arrange: Add commands from different sources sshManager.addBrowserCommand(testSessionName, "user command", "cmd-1", "user"); sshManager.addBrowserCommand(testSessionName, "claude command", "cmd-2", "claude"); sshManager.addBrowserCommand(testSessionName, "another user command", "cmd-3", "user"); // Act: Try to execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "intercepted" }); // Assert: Should include all buffered commands regardless of source expect(result).toEqual({ success: false, error: "BROWSER_COMMANDS_EXECUTED", message: "User executed commands directly in browser", browserCommands: [ { command: "user command", commandId: "cmd-1", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } }, { command: "claude command", commandId: "cmd-2", timestamp: expect.any(Number), source: "claude", result: { stdout: "", stderr: "", exitCode: -1 } }, { command: "another user command", commandId: "cmd-3", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } } ], retryAllowed: true }); }); }); describe("Buffer Clearing After Error Response", () => { it("should clear buffer after sending error response", async () => { // Arrange: Add commands to buffer sshManager.addBrowserCommand(testSessionName, "test command 1", "cmd-1", "user"); sshManager.addBrowserCommand(testSessionName, "test command 2", "cmd-2", "user"); // Verify buffer has content before interception expect(sshManager.getBrowserCommandBuffer(testSessionName)).toHaveLength(2); // Act: Execute command that triggers interception const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "blocked command" }); // Assert: Should get error response expect(result).toMatchObject({ success: false, error: "BROWSER_COMMANDS_EXECUTED" }); // Assert: Buffer should be cleared after error response expect(sshManager.getBrowserCommandBuffer(testSessionName)).toHaveLength(0); }); it("should clear buffer only after error response is sent", async () => { // Arrange: Add commands to buffer sshManager.addBrowserCommand(testSessionName, "buffered command", "cmd-1", "user"); // Verify buffer has content before interception expect(sshManager.getBrowserCommandBuffer(testSessionName)).toHaveLength(1); // Act: Execute command that triggers interception const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "intercepted" }); // Assert: Error response received and buffer cleared expect(result).toMatchObject({ success: false, error: "BROWSER_COMMANDS_EXECUTED" }); // Assert: Buffer should be cleared after error response expect(sshManager.getBrowserCommandBuffer(testSessionName)).toHaveLength(0); }); }); describe("Error Response Format Validation", () => { it("should return properly formatted CommandGatingError", async () => { // Arrange: Add command to buffer sshManager.addBrowserCommand(testSessionName, "format test", "cmd-1", "user"); // Act: Execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "test" }); // Assert: Verify exact error response format expect(result).toEqual({ success: false, error: "BROWSER_COMMANDS_EXECUTED", message: "User executed commands directly in browser", browserCommands: [ { command: "format test", commandId: "cmd-1", timestamp: expect.any(Number), source: "user", result: { stdout: "", stderr: "", exitCode: -1 } } ], retryAllowed: true }); // Assert: Response should be JSON serializable expect(() => JSON.stringify(result)).not.toThrow(); }); it("should maintain JSON-RPC 2.0 compliance in error response", async () => { // Arrange: Add commands to buffer sshManager.addBrowserCommand(testSessionName, "compliance test", "cmd-1", "user"); // Act: Execute command via MCP const result = await mcpServer.callTool("ssh_exec", { sessionName: testSessionName, command: "test" }); // Assert: Should have required fields for structured error expect(result).toHaveProperty("success", false); expect(result).toHaveProperty("error", "BROWSER_COMMANDS_EXECUTED"); expect(result).toHaveProperty("message"); expect(result).toHaveProperty("browserCommands"); expect(result).toHaveProperty("retryAllowed", true); }); }); });