Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Nested swiss-knife reference for Zhipu / Z.AI coding-plan capabilities. Use the Zhipu / Z.AI GLM coding-plan subscription as a unified backend for vision (multimodal screenshot/diagram analysis), web search, web page reading, and zread (open-source GitHub repo browsing). One API key unlocks all four MCP servers. This skill is a thin pointer: it tells you how to source the key, pick the region (Z.AI international vs BigModel mainland), and where the live docs are. MCP server registration is owned
.claude/skills/lingtai-ai-zhipu-coding-plan/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 132% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 163% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 111% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 172% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 51% | 0% |
> Thin pointer. Live docs are the source of truth — curl them when you need depth.
When you need details — current models, exact tool parameters, per-tier quotas, MCP package or endpoint changes — fetch these. Both regions publish equivalent content; pick the one matching the user's account. URLs verified live 2026-04-29.
| Topic | International (Z.AI) | Mainland (BigModel / 智谱) | |---|---|---| | Coding-plan overview | docs.z.ai/devpack/overview | docs.bigmodel.cn/cn/coding-plan/overview | | Vision MCP server | docs.z.ai/devpack/mcp/vision-mcp-server | docs.bigmodel.cn/cn/coding-plan/mcp/vision-mcp-server | | Web search MCP | docs.z.ai/devpack/mcp/search-mcp-server | docs.bigmodel.cn/cn/coding-plan/mcp/search-mcp-server | | Web reader MCP | docs.z.ai/devpack/mcp/reader-mcp-server | docs.bigmodel.cn/cn/coding-plan/mcp/reader-mcp-server | | Zread (GitHub repos) MCP | docs.z.ai/devpack/mcp/zread-mcp-server | docs.bigmodel.cn/cn/coding-plan/mcp/zread-mcp-server | | Doc index (use if any URL above stales) | docs.z.ai/llms.txt | docs.bigmodel.cn/llms.txt |
Note the URL pattern asymmetry: international is /devpack/mcp/..., mainland is /cn/coding-plan/mcp/.... If a path 404s, fetch the llms.txt index above to find the new path — they shift.
Always curl (or use web_search(action="manual")) when you need fresh info — the skill snapshot will go stale.
A single Z.AI / BigModel coding-plan key unlocks four MCP servers. Per-tier quotas (snapshot — verify against the live page):
| Capability | Lite | Pro | Max | |---|---|---|---| | Vision MCP (8 tools — see below) | shared | shared | shared | | Web search (webSearchPrime) | 100 / total | 1,000 / total | 4,000 / total | | Web reader (webReader) | 100 / total | 1,000 / total | 4,000 / total | | Zread (3 tools) | 100 / total | 1,000 / total | 4,000 / total |
Quotas are cumulative-total, not daily — once exhausted, you're done until the plan period rolls over.
| Modality | Server | Tools | |---|---|---| | vision | stdio: npx -y @z_ai/mcp-server | image_analysis, extract_text_from_screenshot, ui_to_artifact, diagnose_error_screenshot, understand_technical_diagram, analyze_data_visualization, ui_diff_check, video_analysis | | web search | http: https://api.z.ai/api/mcp/web_search_prime/mcp | webSearchPrime | | web reader | http: https://api.z.ai/api/mcp/web_reader/mcp | webReader | | zread | http: https://api.z.ai/api/mcp/zread/mcp | search_doc, get_repo_structure, read_file |
Mainland (BigModel) HTTP MCP host is https://open.bigmodel.cn/api/mcp/... — substitute that origin for https://api.z.ai in the URLs above. (api.bigmodel.cn 301-redirects to open.bigmodel.cn.) Verified live 2026-04-29.
MCP server registration is owned by the mcp-manual skill (kernel mcp capability) — read it for how to register, activate, and troubleshoot MCP servers (registry route or legacy mcp/servers.json). This skill assumes the tools are already in your tool list. If the tool you need is not in your list, that's the signal to register the server first.
Never hardcode the key into mcp/servers.json or any committed file. The env block (for stdio) and headers block (for http) are plain text — leak risk on commit, on backup, on screen-share.
Resolution order:
~/.lingtai-tui/.env — ZHIPU_API_KEY=…. The TUI populates this on firstrun.bashgrep -E '^ZHIPU_API_KEY=' ~/.lingtai-tui/.env | cut -d= -f2- | tr -d ' '
The vision MCP also wants a Z_AI_MODE env var (ZAI for international, ZHIPU for mainland) — derive it from the region detection below.
Two ecosystems, not interchangeable — a key from one region returns auth errors against the other host.
| Region | Console | API host (HTTP MCP) | Z_AI_MODE | |---|---|---|---| | International (Z.AI) | z.ai/manage-apikey/apikey-list | api.z.ai | ZAI | | Mainland (BigModel / 智谱) | open.bigmodel.cn | open.bigmodel.cn | ZHIPU |
The MCP server registration must match the region of the key being used.
Auto-detect from the preset library. Walk presets in ~/.lingtai-tui/presets/ (json/jsonc); for each manifest.llm.provider == "zhipu", read manifest.llm.base_url and map it: z.ai/api.z.ai → International/ZAI, bigmodel.cn → Mainland/ZHIPU.
If presets exist for both regions, the user has accounts in both — pick the one matching the key in ~/.lingtai-tui/.env, or ask the user. If no Zhipu preset exists or the result is ambiguous, ask the user. Do not guess.
| Want to … | Use | |---|---| | Analyze image/screenshot/UI/diagram, no vision-capable LLM | This skill — vision MCP (8 specialized tools) | | Search the web with rich snippets | This skill — webSearchPrime (or built-in web_search capability if Zhipu is your LLM provider — that path uses the same MCP) | | Fetch and read a URL | This skill — webReader (or built-in web_search / web-search-manual for non-Zhipu free tier) | | Browse a GitHub repo (search docs, list files, read code) | This skill — zread | | Process a video file (≤8 MB) | This skill — video_analysis | | Pure compute / code generation | Just use your LLM directly |
The vision server exposes 8 tools, more specialized than MiniMax's single understand_image. Pick the right one:
| Want to … | Tool | |---|---| | General "what's in this image?" | image_analysis | | OCR (extract text from screenshot) | extract_text_from_screenshot | | Convert UI mockup → code/spec | ui_to_artifact | | Debug an error screenshot | diagnose_error_screenshot | | Read an architecture / system diagram | understand_technical_diagram | | Read a chart / dashboard | analyze_data_visualization | | Diff two UI screenshots | ui_diff_check | | Analyze a video file (≤8 MB MP4/MOV/M4V) | video_analysis |
Using the right specialized tool gives noticeably better output than image_analysis for charts, diagrams, or OCR.
| Symptom | Likely cause | Fix | |---|---|---| | Tool not in your list | MCP server not registered | Use mcp-manual skill | | 401 Unauthorized / Invalid API key | Region mismatch, or stale key | Verify host matches account region; refresh ~/.lingtai-tui/.env | | 429 Rate limited or quota error | Plan total exhausted | Live docs — check current quota; the cumulative-total clock resets per plan period | | Vision MCP times out on video | File >8 MB or wrong format | Compress with ffmpeg or use a different format (MP4/MOV/M4V only) | | Z_AI_MODE not set warnings | Missing env var on stdio MCP | Set ZAI (international) or ZHIPU (mainland) in the server's env block |
If the live docs contradict this skill, trust the docs and (if the human consents) update this file. The four MCP server URLs and the package name (@z_ai/mcp-server) are the highest-churn items — verify them against docs.z.ai/llms.txt first.
> Found a bug or issue? If you encounter any problems with this skill, load the lingtai-issue-report skill and follow its instructions to report it.
Other measured skills in the registry, with their headline benchmark lift.