DeepSeekTUI.wiki

Tools & capabilities

DeepSeek TUI exposes structured tools—not just chat responses—so the model can act on your repo with guardrails that depend on your mode.

Why tools matter

A coding agent needs more than prose: it must read files, patch them, run tests, inspect git state, and sometimes reach outward for documentation or MCP-connected systems. DeepSeek TUI wires those capabilities into one terminal workflow.

Major tool groups

  • File ops: read, write, patch, list directories.
  • Search: ripgrep-style scans, filename search, structured lookups.
  • Shell: run project commands with approvals (outside Plan mode).
  • Git & diagnostics: understand diffs, commits, and failures.
  • Web: fetch docs or pages when local sources are not enough.
  • RLM helpers: break large tasks into parallel-friendly chunks.
  • MCP: optional external servers—databases, browsers, ticketing systems.

File tools

File tools sit at the center of everyday coding. In Plan mode they are read-only; in Agent mode they can propose edits that you approve; in YOLO they may execute immediately—another reason to avoid YOLO on production trees.

Shell tools

Shell access lets the agent run builds, linters, migrations, or targeted scripts. Treat shell proposals like code review items: verify paths, environment, and side effects before accepting.

Git-aware workflows

Git integration keeps the agent grounded in branch reality—ideal for summarizing changes before you push or generating review notes from local diffs.

Extending with MCP

MCP connects trusted external processes. Once configured, MCP tools appear alongside native ones. Walk through setup on MCP & skills.

FAQ

Can DeepSeek TUI edit multiple files?

Yes—watch approvals closely when changes span many paths.

Does web access replace local docs?

No—prefer local sources when possible to reduce noise and tokens.

Next page?

Read RLM workflows for decomposition patterns.