jkspeaks/claude-code

Purpose and entry points

This repository archives the TypeScript source of Claude Code and adds exploration tooling around it. The primary product is a Bun-based terminal coding agent; companion surfaces include an MCP source-explorer server and a Next.js web UI.

• CLI/bootstrap: src/main.tsx, src/entrypoints/cli.tsx, src/entrypoints/init.ts

• MCP explorer executable/server: mcp-server/src/index.ts, mcp-server/src/server.ts

• Web surface: web/app/page.tsx, web/app/api/chat/route.ts

Interactive CLI and application state

The CLI parses invocation modes, initializes session/runtime dependencies, then renders a React terminal application using the repository’s Ink implementation. Application state, contexts, hooks, screens, commands, and components form the interactive control plane.

• Top-level UI and REPL: src/components/App.tsx, src/screens/REPL.tsx, src/replLauncher.tsx

• State and shared contexts: src/state/AppState.tsx, src/state/AppStateStore.ts, src/context.ts

• Custom terminal renderer: src/ink/ink.tsx, src/ink/renderer.ts, src/ink/layout/engine.ts

Model query pipeline

QueryEngine is the central agent loop: it builds model context, calls Claude-facing API services, handles streaming responses, coordinates tool calls/results, transitions, token budgets, retries, and session progression. Conversation history, compaction, memory extraction, and cost tracking constrain and preserve the session.

• Engine and query mechanics: src/QueryEngine.ts, src/query/transitions.ts, src/query/tokenBudget.ts

• API boundary: src/services/api/claude.ts, src/services/api/client.ts

• Context lifecycle: src/context.ts, src/services/compact/compact.ts, src/services/SessionMemory/sessionMemory.ts

Tools, safety, and execution

The model operates through a registered tool system spanning local files, shell/PowerShell, search, web access, LSP, MCP, tasks, and subagents. Tool orchestration is separated from permission decisions and concrete execution. Permission rules, path/shell validation, UI approvals, and sandbox support are the critical safety boundary before effects occur.

• Tool abstractions/registry: src/Tool.ts, src/tools.ts, src/services/tools/toolOrchestration.ts

• Representative effectful tools: src/tools/BashTool/BashTool.tsx, src/tools/FileEditTool/FileEditTool.ts, src/tools/AgentTool/AgentTool.tsx

• Authorization/sandbox boundary: src/hooks/toolPermission/PermissionContext.ts, src/utils/permissions/permissions.ts, src/utils/sandbox/sandbox-adapter.ts

Commands and extensibility

Slash commands provide user-directed operations separate from model tool calls: session control, configuration, authentication, review, MCP, plugins, tasks, and memory. Plugins, skills, and user-defined agents extend available commands, prompts, tools, output styles, hooks, and integrations.

• Command registry and implementations: src/commands.ts, src/commands/mcp/mcp.tsx, src/commands/plugin/plugin.tsx

• Plugin loading/install lifecycle: src/services/plugins/PluginInstallationManager.ts, src/utils/plugins/pluginLoader.ts

• Skills and agents: src/skills/loadSkillsDir.ts, src/tools/AgentTool/loadAgentsDir.ts

Remote, IDE, and multi-agent boundaries

The bridge subsystem runs bidirectional CLI–IDE sessions, including transport, JWT/trusted-device handling, inbound messages, and permission callbacks. Remote sessions and agent/team tasks provide alternate execution locations and asynchronous work while preserving task/session state.

• IDE bridge: src/bridge/bridgeMain.ts, src/bridge/bridgeMessaging.ts, src/bridge/bridgePermissionCallbacks.ts

• Remote control/session transport: src/remote/RemoteSessionManager.ts, src/remote/SessionsWebSocket.ts

• Task/teammate implementations: src/tasks/LocalMainSessionTask.ts, src/tasks/RemoteAgentTask/RemoteAgentTask.tsx, src/utils/swarm/spawnInProcess.ts

External protocols, runtime, and deployment

Core runtime is Bun with strict TypeScript, React, and Ink; Zod schemas validate configuration/protocol inputs. External integrations include Anthropic APIs, OAuth/JWT and secure storage, MCP servers, LSP servers, GrowthBook/analytics, OpenTelemetry-style telemetry, GitHub, and optional cloud/remote services. Docker artifacts support container deployment; MCP server deployment includes Vercel and Railway-oriented files.

• Protocol services: src/services/mcp/MCPConnectionManager.tsx, src/services/lsp/LSPServerManager.ts, src/services/oauth/client.ts

• Telemetry/flags: src/services/analytics/growthbook.ts, src/utils/telemetry/instrumentation.ts

• Deployment: Dockerfile, docker/docker-compose.yml, mcp-server/api/index.ts