AgentSkillsCN

vector-search-workflows

利用 MCP 向量搜索，构建向量搜索索引与查询工作流，涵盖设置、重新索引、自动索引策略，以及与 MCP 的深度集成。

SKILL.md
--- frontmatter
name: vector-search-workflows
description: Vector search indexing and querying workflows using MCP Vector Search, including setup, reindexing, auto-index strategies, and MCP integration.
version: 1.0.0
category: toolchain
author: Claude MPM Team
license: MIT
progressive_disclosure:
  entry_point:
    summary: "Index a codebase with mcp-vector-search, keep it fresh with auto-indexing, and query via CLI or MCP integration."
    when_to_use: "Building semantic search for codebases, setting up MCP search tools, or troubleshooting indexing and reindexing workflows."
    quick_start: "1. mcp-vector-search setup 2. mcp-vector-search search \"query\" 3. mcp-vector-search index --force when schema changes"
tags:
  - vector-search
  - embeddings
  - indexing
  - search
  - mcp

Vector Search Workflows (MCP Vector Search)

Overview

Use mcp-vector-search to index codebases into ChromaDB and search via semantic embeddings. The recommended flow is setup (init + index + MCP integration), then search, and use index or auto-index to keep data fresh.

Quick Start

bash
pip install mcp-vector-search
mcp-vector-search setup
mcp-vector-search search "authentication logic"

setup detects languages, initializes config, indexes the repo, and configures MCP integrations (Claude Code, Cursor, etc.).

Core Commands

Indexing

bash
mcp-vector-search index
mcp-vector-search index --force
mcp-vector-search index reindex --all --force
mcp-vector-search index reindex path/to/file.py

Auto-Index Strategies

bash
mcp-vector-search auto-index setup --method all
mcp-vector-search auto-index status
mcp-vector-search auto-index check --auto-reindex --max-files 10
mcp-vector-search auto-index teardown --method all

Search

bash
mcp-vector-search search "error handling patterns"
mcp-vector-search search "vector store initialization"

Status + Doctor

bash
mcp-vector-search status
mcp-vector-search doctor

MCP Integration Pattern

setup uses native claude mcp add when available, otherwise falls back to .mcp.json.

Typical .mcp.json entry:

json
{
  "mcpServers": {
    "mcp-vector-search": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "mcp-vector-search", "mcp"],
      "env": {
        "MCP_ENABLE_FILE_WATCHING": "true"
      }
    }
  }
}

Reindex Triggers

  • •Dependency updates or parser changes
  • •Large refactors
  • •Adding new languages or file extensions
  • •Tool upgrades (version tracking triggers reindex)

Local Patterns

  • •Use uv for dev installs: uv sync --dev
  • •Use setup --force to rebuild config + index after tool upgrades
  • •Keep file watching on via MCP_ENABLE_FILE_WATCHING=true

Related Skills

  • •toolchains/ai/protocols/mcp
  • •universal/main/mcp-builder