I Built a Deep Research Skill Using z.ai Coding Plan
Background: Use It or Lose It
I have a z.ai Coding Plan subscription. Honestly? It’s been too unstable to use as an actual coding agent. I’ve been mostly using GPT Pro these days, which made z.ai even more of a shelf warmer.
Then, while tinkering with OpenClaw agents, a thought crossed my mind:
“If I want to run deep research from an OpenClaw agent… couldn’t I just hook up z.ai’s MCP servers?”
I’m paying for the subscription anyway, so I might as well get something out of it. That’s how this skill was born — a deep research skill powered by the four MCP servers that come with z.ai Coding Plan.
What is zai-deep-research?
It’s a deep research skill compatible with the Agent Skills specification.
It relies on four MCP servers provided by z.ai Coding Plan:
| MCP Server | Purpose |
|---|---|
web-search-zai | Web search |
web-reader-zai | Web page content reading |
vision-zai | Image/screenshot analysis |
zread | Document reading |
It’s not tied to any specific coding agent. The bundled Python launcher currently supports codex, claude, opencode, and gemini as backends.
How Does It Work?
It runs a four-stage prompt pipeline:
- Planner: Organizes the request, determines if clarification is needed, and selects which MCPs to use
- Researcher: Iteratively gathers evidence through z.ai MCP servers
- Summarizer: Distills each research round and suggests next queries
- Synthesizer: Produces the final markdown report
Optionally, it supports FAISS-based vector memory for semantic recall across research sessions.
Quick Start
Install
# Clone and install interactively
sh zai-deep-research/scripts/install.sh --source-dir ./zai-deep-research --scope user
# Or one-liner
curl -fsSL https://raw.githubusercontent.com/studiojin-dev/zai-deep-research-skill/main/zai-deep-research/scripts/install.sh | sh -s -- --scope user
Validate
python zai-deep-research/scripts/run.py --validate --client codex
See the README for JSON output, eval harness, vector memory, and more.
Why I Built This
I should be focusing more on making money, but when you start playing with agents… you end up building stuff like this. 😅
But seriously, the reasons are simple:
- Sunk cost recovery: z.ai is too flaky as a coding agent, so at least let me use the MCP servers
- Deep research needs tools: Agents like OpenClaw need search/reading capabilities for web research, and z.ai’s MCPs fit perfectly
- Portability matters: I wanted an Agent Skills-compatible skill, not something locked to one client
In the end, I turned an underused subscription into a reusable, open-source tool.
Repo
If you have a z.ai Coding Plan, give it a spin and let me know what you think.
Related Posts
Note: I’m a solo developer based in Korea. To share my journey with a wider audience, I used AI to help translate my thoughts into English. If any phrasing feels a bit “too AI” or unnatural, please bear with me—I’m spending most of my energy on the code!