AgentSkillsCN

mcp-setup

为Claude Code、Cursor或Windsurf配置GrepAI MCP服务器，灵活设置作用域与工作空间选项。

SKILL.md
--- frontmatter
name: mcp-setup
description: Configure grepai MCP server for Claude Code, Cursor, or Windsurf with scope and workspace options.
allowed-tools:
  - Read
  - Glob
  - Grep
  - Bash(claude *)
  - Bash(which *)
  - Bash(cat *)
  - Write
  - Edit
  - AskUserQuestion

GrepAI MCP Server Configuration

Configure grepai as an MCP server for AI coding assistants. Supports multiple IDEs, scopes, and workspace modes.

Workflow

1. Detect Current State

Check all possible MCP registration locations and grepai config:

bash
# Check existing MCP registrations
claude mcp list 2>/dev/null
text
Glob: .mcp.json
Glob: .claude/mcp.json
Read: ~/.claude.json  (look for mcpServers section)

Check for workspaces:

bash
grepai workspace list 2>/dev/null

Check for local project config:

text
Glob: .grepai/config.yaml

Summarize findings before proceeding:

  • •Existing MCP registrations (if any)
  • •Available workspaces (if any)
  • •Local project config (if any)
  • •grepai binary location: which grepai

2. Ask: IDE Target

Ask via AskUserQuestion:

text
Which IDE to configure?

○ Claude Code (Recommended)
○ Cursor
○ Windsurf
○ Generic .mcp.json

3. Ask: Registration Scope

For Claude Code, ask via AskUserQuestion:

text
Where should the MCP server be registered?

○ Project .mcp.json (Recommended for teams — shareable via git)
○ User global (~/.claude.json — available in all sessions)
○ Project .claude/mcp.json (this project only, not shared)

For Cursor: config goes to .cursor/mcp.json (project-level).

For Windsurf: config goes to .windsurf/mcp.json (project-level).

For Generic: config goes to .mcp.json at project root.

4. Ask: Workspace Mode

Only ask this if workspaces were detected in step 1.

Ask via AskUserQuestion:

text
Workspace mode?

○ With --workspace {NAME} (Recommended — auto-searches all projects without extra params)
○ Without workspace (agent must pass workspace parameter manually)
○ No workspace (single project mode)

If workspace selected, show the available workspaces and let user pick:

text
Which workspace?

○ {ws1} ({N} projects)
○ {ws2} ({N} projects)

5. Generate Configuration

Determine the grepai binary path:

bash
which grepai

For Claude Code — claude mcp add

User scope:

bash
claude mcp add grepai -s user -- grepai mcp-serve {ARGS}

Project scope:

bash
claude mcp add grepai -s project -- grepai mcp-serve {ARGS}

Where {ARGS} is:

  • •No workspace: empty (auto-detects from cwd)
  • •With workspace: --workspace {NAME}
  • •With explicit path: {PROJECT_PATH}

For .mcp.json (project root, Cursor, Windsurf, Generic)

Determine the target file:

  • •Cursor: .cursor/mcp.json
  • •Windsurf: .windsurf/mcp.json
  • •Generic / Project .mcp.json: .mcp.json

Write the JSON config:

json
{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve"]
    }
  }
}

With workspace:

json
{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve", "--workspace", "{NAME}"]
    }
  }
}

With explicit project path:

json
{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve", "{PROJECT_PATH}"]
    }
  }
}

If the target JSON file already exists, read it first and merge the grepai key into the existing mcpServers object. Do not overwrite other MCP servers.

Create parent directories if needed

bash
mkdir -p .cursor   # for Cursor
mkdir -p .windsurf # for Windsurf

6. Add to .gitignore (if applicable)

For project-level .mcp.json files that teams share, do NOT gitignore them.

For .claude/mcp.json (project-specific, not shared), check if .claude/ is in .gitignore. If not, suggest adding it.

7. Verify

For Claude Code:

bash
claude mcp list

Confirm grepai appears in the list.

For file-based configs:

Read the written file to confirm it's valid JSON.

Note: The MCP server connects on next IDE session start, not immediately.

8. Print Summary

text
============================================================================
GrepAI MCP Server Configured
============================================================================

IDE:        {IDE}
Scope:      {SCOPE}
Config:     {FILE_PATH}
Command:    grepai mcp-serve {ARGS}
Workspace:  {NAME or "none"}

The MCP server will be available in your next {IDE} session.
It exposes these tools:

  grepai_search         — Semantic code search
  grepai_trace_callers  — Find all callers of a function
  grepai_trace_callees  — Find all callees of a function
  grepai_trace_graph    — Build call graph around a symbol
  grepai_index_status   — Check index health

{IF WORKSPACE}
Workspace mode is enabled. The agent will automatically search
across all {N} projects in the "{NAME}" workspace without needing
to specify the workspace parameter.
{END IF}

To verify after restart:
  claude mcp list       # Claude Code
============================================================================