cli/pkg/mcp/tools_process.go

302 lines
9.5 KiB
Go
Raw Normal View History

feat: BugSETI app, WebSocket hub, browser automation, and MCP tools (#336) * feat: add security logging and fix framework regressions This commit implements comprehensive security event logging and resolves critical regressions in the core framework. Security Logging: - Enhanced `pkg/log` with a `Security` level and helper. - Added `log.Username()` to consistently identify the executing user. - Instrumented GitHub CLI auth, Agentic configuration, filesystem sandbox, MCP handlers, and MCP TCP transport with security logs. - Added `SecurityStyle` to the CLI for consistent visual representation of security events. UniFi Security (CodeQL): - Refactored `pkg/unifi` to remove hardcoded `InsecureSkipVerify`, resolving a high-severity alert. - Added a `--verify-tls` flag and configuration option to control TLS verification. - Updated command handlers to support the new verification parameter. Framework Fixes: - Restored original signatures for `MustServiceFor`, `Config()`, and `Display()` in `pkg/framework/core`, which had been corrupted during a merge. - Fixed `pkg/framework/framework.go` and `pkg/framework/core/runtime_pkg.go` to match the restored signatures. - These fixes resolve project-wide compilation errors caused by the signature mismatches. I encountered significant blockers due to a corrupted state of the `dev` branch after a merge, which introduced breaking changes in the core framework's DI system. I had to manually reconcile these signatures with the expected usage across the codebase to restore build stability. * feat(mcp): add RAG tools (query, ingest, collections) Add vector database tools to the MCP server for RAG operations: - rag_query: Search for relevant documentation using semantic similarity - rag_ingest: Ingest files or directories into the vector database - rag_collections: List available collections Uses existing internal/cmd/rag exports (QueryDocs, IngestDirectory, IngestFile) and pkg/rag for Qdrant client access. Default collection is "hostuk-docs" with topK=5 for queries. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(mcp): add metrics tools (record, query) Add MCP tools for recording and querying AI/security metrics events. The metrics_record tool writes events to daily JSONL files, and the metrics_query tool provides aggregated statistics by type, repo, and agent. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat: add 'core mcp serve' command Add CLI command to start the MCP server for AI tool integration. - Create internal/cmd/mcpcmd package with serve subcommand - Support --workspace flag for directory restriction - Handle SIGINT/SIGTERM for clean shutdown - Register in full.go build variant Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(ws): add WebSocket hub package for real-time streaming Add pkg/ws package implementing a hub pattern for WebSocket connections: - Hub manages client connections, broadcasts, and channel subscriptions - Client struct represents connected WebSocket clients - Message types: process_output, process_status, event, error, ping/pong - Channel-based subscription system (subscribe/unsubscribe) - SendProcessOutput and SendProcessStatus for process streaming integration - Full test coverage including concurrency tests Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(mcp): add process management and WebSocket MCP tools Add MCP tools for process management: - process_start: Start a new external process - process_stop: Gracefully stop a running process - process_kill: Force kill a process - process_list: List all managed processes - process_output: Get captured process output - process_input: Send input to process stdin Add MCP tools for WebSocket: - ws_start: Start WebSocket server for real-time streaming - ws_info: Get hub statistics (clients, channels) Update Service struct with optional process.Service and ws.Hub fields, new WithProcessService and WithWSHub options, getter methods, and Shutdown method for cleanup. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(webview): add browser automation package via Chrome DevTools Protocol Add pkg/webview package for browser automation: - webview.go: Main interface with Connect, Navigate, Click, Type, QuerySelector, Screenshot, Evaluate - cdp.go: Chrome DevTools Protocol WebSocket client implementation - actions.go: DOM action types (Click, Type, Hover, Scroll, etc.) and ActionSequence builder - console.go: Console message capture and filtering with ConsoleWatcher and ExceptionWatcher - angular.go: Angular-specific helpers for router navigation, component access, and Zone.js stability Add MCP tools for webview: - webview_connect/disconnect: Connection management - webview_navigate: Page navigation - webview_click/type/query/wait: DOM interaction - webview_console: Console output capture - webview_eval: JavaScript execution - webview_screenshot: Screenshot capture Add documentation: - docs/mcp/angular-testing.md: Guide for Angular application testing Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * docs: document new packages and BugSETI application - Update CLAUDE.md with documentation for: - pkg/ws (WebSocket hub for real-time streaming) - pkg/webview (Browser automation via CDP) - pkg/mcp (MCP server tools: process, ws, webview) - BugSETI application overview - Add comprehensive README for BugSETI with: - Installation and configuration guide - Usage workflow documentation - Architecture overview - Contributing guidelines Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(bugseti): add BugSETI system tray app with auto-update BugSETI - Distributed Bug Fixing like SETI@home but for code Features: - System tray app with Wails v3 - GitHub issue fetching with label filters - Issue queue with priority management - AI context seeding via seed-agent-developer skill - Automated PR submission flow - Stats tracking and leaderboard - Cross-platform notifications - Self-updating with stable/beta/nightly channels Includes: - cmd/bugseti: Main application with Angular frontend - internal/bugseti: Core services (fetcher, queue, seeder, submit, config, stats, notify) - internal/bugseti/updater: Auto-update system (checker, downloader, installer) - .github/workflows/bugseti-release.yml: CI/CD for all platforms Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix: resolve import cycle and code duplication - Remove pkg/log import from pkg/io/local to break import cycle (pkg/log/rotation.go imports pkg/io, creating circular dependency) - Use stderr logging for security events in sandbox escape detection - Remove unused sync/atomic import from core.go - Fix duplicate LogSecurity function declarations in cli/log.go - Update workspace/service.go Crypt() call to match interface Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix: update tests for new function signatures and format code - Update core_test.go: Config(), Display() now panic instead of returning error - Update runtime_pkg_test.go: sr.Config() now panics instead of returning error - Update MustServiceFor tests to use assert.Panics - Format BugSETI, MCP tools, and webview packages with gofmt Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> --------- Co-authored-by: Snider <631881+Snider@users.noreply.github.com> Co-authored-by: Claude <developers@lethean.io> Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-05 17:22:05 +00:00
package mcp
import (
"context"
"fmt"
"time"
"forge.lthn.ai/core/cli/pkg/log"
"forge.lthn.ai/core/cli/pkg/process"
feat: BugSETI app, WebSocket hub, browser automation, and MCP tools (#336) * feat: add security logging and fix framework regressions This commit implements comprehensive security event logging and resolves critical regressions in the core framework. Security Logging: - Enhanced `pkg/log` with a `Security` level and helper. - Added `log.Username()` to consistently identify the executing user. - Instrumented GitHub CLI auth, Agentic configuration, filesystem sandbox, MCP handlers, and MCP TCP transport with security logs. - Added `SecurityStyle` to the CLI for consistent visual representation of security events. UniFi Security (CodeQL): - Refactored `pkg/unifi` to remove hardcoded `InsecureSkipVerify`, resolving a high-severity alert. - Added a `--verify-tls` flag and configuration option to control TLS verification. - Updated command handlers to support the new verification parameter. Framework Fixes: - Restored original signatures for `MustServiceFor`, `Config()`, and `Display()` in `pkg/framework/core`, which had been corrupted during a merge. - Fixed `pkg/framework/framework.go` and `pkg/framework/core/runtime_pkg.go` to match the restored signatures. - These fixes resolve project-wide compilation errors caused by the signature mismatches. I encountered significant blockers due to a corrupted state of the `dev` branch after a merge, which introduced breaking changes in the core framework's DI system. I had to manually reconcile these signatures with the expected usage across the codebase to restore build stability. * feat(mcp): add RAG tools (query, ingest, collections) Add vector database tools to the MCP server for RAG operations: - rag_query: Search for relevant documentation using semantic similarity - rag_ingest: Ingest files or directories into the vector database - rag_collections: List available collections Uses existing internal/cmd/rag exports (QueryDocs, IngestDirectory, IngestFile) and pkg/rag for Qdrant client access. Default collection is "hostuk-docs" with topK=5 for queries. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(mcp): add metrics tools (record, query) Add MCP tools for recording and querying AI/security metrics events. The metrics_record tool writes events to daily JSONL files, and the metrics_query tool provides aggregated statistics by type, repo, and agent. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat: add 'core mcp serve' command Add CLI command to start the MCP server for AI tool integration. - Create internal/cmd/mcpcmd package with serve subcommand - Support --workspace flag for directory restriction - Handle SIGINT/SIGTERM for clean shutdown - Register in full.go build variant Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(ws): add WebSocket hub package for real-time streaming Add pkg/ws package implementing a hub pattern for WebSocket connections: - Hub manages client connections, broadcasts, and channel subscriptions - Client struct represents connected WebSocket clients - Message types: process_output, process_status, event, error, ping/pong - Channel-based subscription system (subscribe/unsubscribe) - SendProcessOutput and SendProcessStatus for process streaming integration - Full test coverage including concurrency tests Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(mcp): add process management and WebSocket MCP tools Add MCP tools for process management: - process_start: Start a new external process - process_stop: Gracefully stop a running process - process_kill: Force kill a process - process_list: List all managed processes - process_output: Get captured process output - process_input: Send input to process stdin Add MCP tools for WebSocket: - ws_start: Start WebSocket server for real-time streaming - ws_info: Get hub statistics (clients, channels) Update Service struct with optional process.Service and ws.Hub fields, new WithProcessService and WithWSHub options, getter methods, and Shutdown method for cleanup. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(webview): add browser automation package via Chrome DevTools Protocol Add pkg/webview package for browser automation: - webview.go: Main interface with Connect, Navigate, Click, Type, QuerySelector, Screenshot, Evaluate - cdp.go: Chrome DevTools Protocol WebSocket client implementation - actions.go: DOM action types (Click, Type, Hover, Scroll, etc.) and ActionSequence builder - console.go: Console message capture and filtering with ConsoleWatcher and ExceptionWatcher - angular.go: Angular-specific helpers for router navigation, component access, and Zone.js stability Add MCP tools for webview: - webview_connect/disconnect: Connection management - webview_navigate: Page navigation - webview_click/type/query/wait: DOM interaction - webview_console: Console output capture - webview_eval: JavaScript execution - webview_screenshot: Screenshot capture Add documentation: - docs/mcp/angular-testing.md: Guide for Angular application testing Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * docs: document new packages and BugSETI application - Update CLAUDE.md with documentation for: - pkg/ws (WebSocket hub for real-time streaming) - pkg/webview (Browser automation via CDP) - pkg/mcp (MCP server tools: process, ws, webview) - BugSETI application overview - Add comprehensive README for BugSETI with: - Installation and configuration guide - Usage workflow documentation - Architecture overview - Contributing guidelines Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * feat(bugseti): add BugSETI system tray app with auto-update BugSETI - Distributed Bug Fixing like SETI@home but for code Features: - System tray app with Wails v3 - GitHub issue fetching with label filters - Issue queue with priority management - AI context seeding via seed-agent-developer skill - Automated PR submission flow - Stats tracking and leaderboard - Cross-platform notifications - Self-updating with stable/beta/nightly channels Includes: - cmd/bugseti: Main application with Angular frontend - internal/bugseti: Core services (fetcher, queue, seeder, submit, config, stats, notify) - internal/bugseti/updater: Auto-update system (checker, downloader, installer) - .github/workflows/bugseti-release.yml: CI/CD for all platforms Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix: resolve import cycle and code duplication - Remove pkg/log import from pkg/io/local to break import cycle (pkg/log/rotation.go imports pkg/io, creating circular dependency) - Use stderr logging for security events in sandbox escape detection - Remove unused sync/atomic import from core.go - Fix duplicate LogSecurity function declarations in cli/log.go - Update workspace/service.go Crypt() call to match interface Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix: update tests for new function signatures and format code - Update core_test.go: Config(), Display() now panic instead of returning error - Update runtime_pkg_test.go: sr.Config() now panics instead of returning error - Update MustServiceFor tests to use assert.Panics - Format BugSETI, MCP tools, and webview packages with gofmt Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> --------- Co-authored-by: Snider <631881+Snider@users.noreply.github.com> Co-authored-by: Claude <developers@lethean.io> Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-05 17:22:05 +00:00
"github.com/modelcontextprotocol/go-sdk/mcp"
)
// ProcessStartInput contains parameters for starting a new process.
type ProcessStartInput struct {
Command string `json:"command"` // The command to run
Args []string `json:"args,omitempty"` // Command arguments
Dir string `json:"dir,omitempty"` // Working directory
Env []string `json:"env,omitempty"` // Environment variables (KEY=VALUE format)
}
// ProcessStartOutput contains the result of starting a process.
type ProcessStartOutput struct {
ID string `json:"id"`
PID int `json:"pid"`
Command string `json:"command"`
Args []string `json:"args"`
StartedAt time.Time `json:"startedAt"`
}
// ProcessStopInput contains parameters for gracefully stopping a process.
type ProcessStopInput struct {
ID string `json:"id"` // Process ID to stop
}
// ProcessStopOutput contains the result of stopping a process.
type ProcessStopOutput struct {
ID string `json:"id"`
Success bool `json:"success"`
Message string `json:"message,omitempty"`
}
// ProcessKillInput contains parameters for force killing a process.
type ProcessKillInput struct {
ID string `json:"id"` // Process ID to kill
}
// ProcessKillOutput contains the result of killing a process.
type ProcessKillOutput struct {
ID string `json:"id"`
Success bool `json:"success"`
Message string `json:"message,omitempty"`
}
// ProcessListInput contains parameters for listing processes.
type ProcessListInput struct {
RunningOnly bool `json:"running_only,omitempty"` // If true, only return running processes
}
// ProcessListOutput contains the list of processes.
type ProcessListOutput struct {
Processes []ProcessInfo `json:"processes"`
Total int `json:"total"`
}
// ProcessInfo represents information about a process.
type ProcessInfo struct {
ID string `json:"id"`
Command string `json:"command"`
Args []string `json:"args"`
Dir string `json:"dir"`
Status string `json:"status"`
PID int `json:"pid"`
ExitCode int `json:"exitCode"`
StartedAt time.Time `json:"startedAt"`
Duration time.Duration `json:"duration"`
}
// ProcessOutputInput contains parameters for getting process output.
type ProcessOutputInput struct {
ID string `json:"id"` // Process ID
}
// ProcessOutputOutput contains the captured output of a process.
type ProcessOutputOutput struct {
ID string `json:"id"`
Output string `json:"output"`
}
// ProcessInputInput contains parameters for sending input to a process.
type ProcessInputInput struct {
ID string `json:"id"` // Process ID
Input string `json:"input"` // Input to send to stdin
}
// ProcessInputOutput contains the result of sending input to a process.
type ProcessInputOutput struct {
ID string `json:"id"`
Success bool `json:"success"`
Message string `json:"message,omitempty"`
}
// registerProcessTools adds process management tools to the MCP server.
// Returns false if process service is not available.
func (s *Service) registerProcessTools(server *mcp.Server) bool {
if s.processService == nil {
return false
}
mcp.AddTool(server, &mcp.Tool{
Name: "process_start",
Description: "Start a new external process. Returns process ID for tracking.",
}, s.processStart)
mcp.AddTool(server, &mcp.Tool{
Name: "process_stop",
Description: "Gracefully stop a running process by ID.",
}, s.processStop)
mcp.AddTool(server, &mcp.Tool{
Name: "process_kill",
Description: "Force kill a process by ID. Use when process_stop doesn't work.",
}, s.processKill)
mcp.AddTool(server, &mcp.Tool{
Name: "process_list",
Description: "List all managed processes. Use running_only=true for only active processes.",
}, s.processList)
mcp.AddTool(server, &mcp.Tool{
Name: "process_output",
Description: "Get the captured output of a process by ID.",
}, s.processOutput)
mcp.AddTool(server, &mcp.Tool{
Name: "process_input",
Description: "Send input to a running process stdin.",
}, s.processInput)
return true
}
// processStart handles the process_start tool call.
func (s *Service) processStart(ctx context.Context, req *mcp.CallToolRequest, input ProcessStartInput) (*mcp.CallToolResult, ProcessStartOutput, error) {
s.logger.Security("MCP tool execution", "tool", "process_start", "command", input.Command, "args", input.Args, "dir", input.Dir, "user", log.Username())
if input.Command == "" {
return nil, ProcessStartOutput{}, fmt.Errorf("command cannot be empty")
}
opts := process.RunOptions{
Command: input.Command,
Args: input.Args,
Dir: input.Dir,
Env: input.Env,
}
proc, err := s.processService.StartWithOptions(ctx, opts)
if err != nil {
log.Error("mcp: process start failed", "command", input.Command, "err", err)
return nil, ProcessStartOutput{}, fmt.Errorf("failed to start process: %w", err)
}
info := proc.Info()
return nil, ProcessStartOutput{
ID: proc.ID,
PID: info.PID,
Command: proc.Command,
Args: proc.Args,
StartedAt: proc.StartedAt,
}, nil
}
// processStop handles the process_stop tool call.
func (s *Service) processStop(ctx context.Context, req *mcp.CallToolRequest, input ProcessStopInput) (*mcp.CallToolResult, ProcessStopOutput, error) {
s.logger.Security("MCP tool execution", "tool", "process_stop", "id", input.ID, "user", log.Username())
if input.ID == "" {
return nil, ProcessStopOutput{}, fmt.Errorf("id cannot be empty")
}
proc, err := s.processService.Get(input.ID)
if err != nil {
log.Error("mcp: process stop failed", "id", input.ID, "err", err)
return nil, ProcessStopOutput{}, fmt.Errorf("process not found: %w", err)
}
// For graceful stop, we use Kill() which sends SIGKILL
// A more sophisticated implementation could use SIGTERM first
if err := proc.Kill(); err != nil {
log.Error("mcp: process stop kill failed", "id", input.ID, "err", err)
return nil, ProcessStopOutput{}, fmt.Errorf("failed to stop process: %w", err)
}
return nil, ProcessStopOutput{
ID: input.ID,
Success: true,
Message: "Process stop signal sent",
}, nil
}
// processKill handles the process_kill tool call.
func (s *Service) processKill(ctx context.Context, req *mcp.CallToolRequest, input ProcessKillInput) (*mcp.CallToolResult, ProcessKillOutput, error) {
s.logger.Security("MCP tool execution", "tool", "process_kill", "id", input.ID, "user", log.Username())
if input.ID == "" {
return nil, ProcessKillOutput{}, fmt.Errorf("id cannot be empty")
}
if err := s.processService.Kill(input.ID); err != nil {
log.Error("mcp: process kill failed", "id", input.ID, "err", err)
return nil, ProcessKillOutput{}, fmt.Errorf("failed to kill process: %w", err)
}
return nil, ProcessKillOutput{
ID: input.ID,
Success: true,
Message: "Process killed",
}, nil
}
// processList handles the process_list tool call.
func (s *Service) processList(ctx context.Context, req *mcp.CallToolRequest, input ProcessListInput) (*mcp.CallToolResult, ProcessListOutput, error) {
s.logger.Info("MCP tool execution", "tool", "process_list", "running_only", input.RunningOnly, "user", log.Username())
var procs []*process.Process
if input.RunningOnly {
procs = s.processService.Running()
} else {
procs = s.processService.List()
}
result := make([]ProcessInfo, len(procs))
for i, p := range procs {
info := p.Info()
result[i] = ProcessInfo{
ID: info.ID,
Command: info.Command,
Args: info.Args,
Dir: info.Dir,
Status: string(info.Status),
PID: info.PID,
ExitCode: info.ExitCode,
StartedAt: info.StartedAt,
Duration: info.Duration,
}
}
return nil, ProcessListOutput{
Processes: result,
Total: len(result),
}, nil
}
// processOutput handles the process_output tool call.
func (s *Service) processOutput(ctx context.Context, req *mcp.CallToolRequest, input ProcessOutputInput) (*mcp.CallToolResult, ProcessOutputOutput, error) {
s.logger.Info("MCP tool execution", "tool", "process_output", "id", input.ID, "user", log.Username())
if input.ID == "" {
return nil, ProcessOutputOutput{}, fmt.Errorf("id cannot be empty")
}
output, err := s.processService.Output(input.ID)
if err != nil {
log.Error("mcp: process output failed", "id", input.ID, "err", err)
return nil, ProcessOutputOutput{}, fmt.Errorf("failed to get process output: %w", err)
}
return nil, ProcessOutputOutput{
ID: input.ID,
Output: output,
}, nil
}
// processInput handles the process_input tool call.
func (s *Service) processInput(ctx context.Context, req *mcp.CallToolRequest, input ProcessInputInput) (*mcp.CallToolResult, ProcessInputOutput, error) {
s.logger.Security("MCP tool execution", "tool", "process_input", "id", input.ID, "user", log.Username())
if input.ID == "" {
return nil, ProcessInputOutput{}, fmt.Errorf("id cannot be empty")
}
if input.Input == "" {
return nil, ProcessInputOutput{}, fmt.Errorf("input cannot be empty")
}
proc, err := s.processService.Get(input.ID)
if err != nil {
log.Error("mcp: process input get failed", "id", input.ID, "err", err)
return nil, ProcessInputOutput{}, fmt.Errorf("process not found: %w", err)
}
if err := proc.SendInput(input.Input); err != nil {
log.Error("mcp: process input send failed", "id", input.ID, "err", err)
return nil, ProcessInputOutput{}, fmt.Errorf("failed to send input: %w", err)
}
return nil, ProcessInputOutput{
ID: input.ID,
Success: true,
Message: "Input sent successfully",
}, nil
}