SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsinstall-mcp

install-mcp

Deployment & operations

View repository

Updated 10-16 07:26

About this skill

install-mcp

Responsibility Boundaries

This skill is the inventory and scheduling entry point for MCP configurations and does not add installation scripts. The execution model should complete preview, backup, and precise writes according to this inventory; this skill does not replace any dedicated installer.

  • The actual installation and verification of Memorix full mode is handed off to init-simple-memorix。
  • Generic third-party MCP is handled only after confirming the server entry schema,command、args、env , and the impact on trust approval.
  • All writes must be premised on preserving existing configurations and must not overwrite other mcpServers。

Known Configuration Targets

Platform Configuration Path Format
Codex ~/.codex/config.toml TOML
Codex ~/.codex/config-2026-6-13-bg.toml TOML
Claude Code ~/.claude.json JSON
Cursor ~/.cursor/mcp.json JSON
WorkBuddy ~/.workbuddy/mcp.json JSON
WorkBuddy ~/.workbuddy/.mcp.json JSON
ZCode ~/.zcode/cli/config.json JSON
Qoder ~/AppData/Roaming/Qoder/SharedClientCache/mcp.json JSON
Qoder ~/.qoder/mcp.json JSON
MiniMax Code ~/.minimax/mcp/mcp.json JSON
Kiro ~/.kiro/settings/mcp.json JSON

Configuration Merge Rules

  • JSON configurations typically store server entries as a mcpServers object; only add or update the target entry, preserving unknown top-level fields and other entries.
  • TOML configurations use their existing MCP server table form; only modify the target table, preserving comments, unknown fields, and other server tables.
  • First parse the current file and generate a dry-run preview, then write after authorization; before the actual write, create an identifiable backup of the original file.
  • If the configuration does not exist, create it only after confirming the minimal file structure accepted by that platform; if configuration parsing fails or the structure is unclear, stop writing and report.
  • Do not overwrite existing configurations wholesale with a template, and do not assume that JSON entries from different platforms are directly interchangeable.

WorkBuddy Special Handling

WorkBuddy may inject Node parameters into child processes. For MCP entries that need to isolate Node parameters, it is recommended to set env.NODE_OPTIONS = "". This is a compatibility recommendation and does not mean that all scripts or all configurations will automatically fill in this field.

Changing WorkBuddy's MCP entry may trigger a new trust approval; after writing, prompt to restart the application and reconfirm the server connection status.

MiniMax Code Special Handling

~/.minimax/mcp/mcp.json is the user-level MCP configuration for the MiniMax Code local agent tool, with a top-level structure of mcpServers, but its entry fields are richer than generic JSON configurations:

  • stdio-form entries use command、args、env; remote-form entries use url + type: "streamable-http", and may carry the headers authentication field.
  • Entries often carry platform metadata fields such as enabled、configured、builtin、description、timeout , which must be preserved as-is during merging and must not be deleted or rewritten.
  • builtin: true 's entry is a built-in service managed by MiniMax Code itself; batch installation only adds or updates non-built-in entries, and does not modify or delete built-in entries.
  • When adding a new entry, preferentially refer to the field forms of existing non-built-in entries in that file, and do not guess the platform's required field set.

This file and the tokens.json in the same directory may contain a Bearer token or API key: dry-run previews, backups, and reports must not output these sensitive fields as-is, nor write them into commits or logs.

Qoder Special Handling

~/.qoder/mcp.json is the user-level MCP configuration purely for the Qoder agent tool, with a top-level structure of mcpServers, and entries in a minimal form (usually only command、args, optionally type). Keep the minimal form when merging, and do not inject metadata fields from other platforms. Note that the target file is mcp.json, not the mcp-router.json。

Qoder-related directory ownership is extremely easy to confuse, so directory ownership must be audited one by one before writing:

  • ~/.qoder-cli、~/.qoder-cn、~/.qoderwork、~/.qoderworkcn are directories for different products; none of them is simply Qoder.
  • Among them, ~/.qoder-cn/mcp.json、~/.qoderworkcn/mcp.json also genuinely exist, but they are not writable targets of this skill; you must not write across them just because the names are similar.
  • In the list, ~/AppData/Roaming/Qoder/SharedClientCache/mcp.json and ~/.qoder/mcp.json are two mutually independent configuration locations; before writing, you must make clear which one this operation targets.
  • ~/.qoder Under the directory there are also extensions/、plugins/、skills/、memories/ and other Qoder runtime directories; batch MCP installation writes only to mcp.jsonand does not touch other directories.

Memorix and third-party MCP orchestration

Scenario Orchestration rule
Memorix full mode Leave it to init-simple-memorix; do not repeat its installation details in this skill.
Known third-party server First verify that server's official entry schema, command, arguments, environment variables, and trust impact, then merge precisely.
Unknown third-party server Do not write; first obtain a reliable configuration source and user authorization.

Future candidates

Antigravity, Trae, Gemini CLI, etc. are only candidate platforms. Only after finding a reliable, verifiable MCP configuration path and its format may they be added to the "writable targets"; you must not hardcode paths based on common naming or historical impressions. In Qoder-like similar directories (such as ~/.qoder-cn、~/.qoderworkcn), existing mcp.json belong to different products and must not be included in writable targets without ownership confirmation and explicit user authorization.

Execution and acceptance

  1. From the known configuration list, select targets that actually exist and are authorized, and confirm the JSON or TOML structure.
  2. Construct a dry-run change that affects only the target server entry, and clearly state the fields that will be preserved, the backup location, and the possible trust approval impact.
  3. After writing, re-parse the configuration to confirm that other mcpServers are unchanged, and check the server status after restarting or refreshing as required by the platform.