Skip to main content
Claude Code ships with a first-class MCP CLI. One command and you’re done.

​
Quick install

claude mcp add sharc -e SHARC_API_KEY=sk_mcp_... -- npx -y @sharc-code/mcp@latest
Restart Claude Code and the seven SHARC tools become available.
Use -s user to install for your user (available across all projects) or -s project to commit a .mcp.json to the repo so teammates get it automatically. Default is local (this project, your machine).

​
Verify

claude mcp list
Look for a sharc entry. From inside Claude Code, ask: 
> List the MCP tools you have available
You should see index_codebase, search_code, clear_index, get_indexing_status, start_watch, stop_watch, get_watch_status.

​
Manual configuration (alternative)

If you’d rather edit JSON directly, add this to your Claude Code MCP config: 
{
  "mcpServers": {
    "sharc": {
      "command": "npx",
      "args": ["-y", "@sharc-code/mcp@latest"],
      "env": {
        "SHARC_API_KEY": "sk_mcp_..."
      }
    }
  }
}

​
Usage examples

​
Indexing a codebase

> Please index this codebase for semantic search

● index_codebase (MCP)
  path: "/Users/you/project"
  ⎿ Scanning files...
  ⎿ Found 342 files to index
  ⎿ Chunking code (AST-based for .ts, .py, .go...)
  ⎿ Generating embeddings (2,450 chunks)
  ⎿ ✓ Indexed 2,450 chunks in 18.3s

​
Searching code

> How does the authentication middleware work?

● search_code (MCP)
  query: "authentication middleware"
  limit: 3
  ⎿ Found 3 results for query: "authentication middleware"

  1. Code snippet (typescript) [project]
     Location: src/middleware/auth.ts:45-89
     Score: 0.9847
     ... +44 lines (ctrl+o to expand)

​
Incremental updates

After the initial index, SHARC watches the working tree and re-indexes changed files automatically: 
> I just modified the auth middleware, is the index updated?

● get_watch_status (MCP)
  ⎿ Watching: /Users/you/project
  ⎿ Last update: 2 seconds ago

​
Best practices

​
Index before deep dives

For the best experience, index before asking complex questions: 
> Before we start, please index this codebase

[After indexing completes]

> Now, explain the data flow from API request to database

​
Use natural language queries

SHARC understands semantic meaning, not just keywords: 
# Good queries:
"How are users authenticated?"
"Where is the database connection configured?"
"Find the error handling for API requests"

# Less effective:
"auth" (too vague)
"config.ts" (use glob patterns for file search)

​
Adjust result limits

For broad exploration, increase the limit: 
> Search for all API endpoints with limit 10
For specific answers, keep it low (default 3).

​
Troubleshooting

​
Tools not available

  1. Restart Claude Code
  2. Run claude mcp list and confirm sharc is listed
  3. Verify the API key is correct: claude mcp get sharc

​
Slow indexing

Initial indexing time depends on codebase size. The Hono repo takes ~30-45 seconds; large repos like Next.js can take 3-4 minutes. Subsequent runs use Merkle diff for near-instant updates.

​
Search returns unexpected results

  • Try rephrasing your query semantically
  • Use extensionFilter to narrow to specific file types
  • Check if the codebase is fully indexed with get_indexing_status