📚 External MCP Integration: Zotero (Local Citation Library)

The research-skills ecosystem comes with a built-in semantic-search provider that hits the public Semantic Scholar API. This is perfect for discovering global papers.

However, for a highly rigorous systematic review, you may only want the AI to query your own carefully vetted local Zotero library. By connecting an external Zotero MCP, the AI will limit its evidence extraction exclusively to papers you have collected and annotated.

How it works (The Fallback Chain)

By default, when Claude or Codex reaches the scholarly-search stage, it routes the request through our MCPConnector.

1. Check Environment Variable: It first looks for RESEARCH_MCP_SCHOLARLY_SEARCH_CMD.
2. Fallback to Native: If not set, it executes our built-in Python script (scripts/mcp_scholarly_search.py).

By setting the environment variable, you override the default API behavior and map the capability directly to your local instance.

Step 1: Install a Zotero MCP Server

You need an MCP server that bridges the Model Context Protocol to your local Zotero database. There are several community options available.

(Example using typical community Zotero MCP):

# Ensure Node.js is installed
npm install -g @zcaceres/zotero-mcp-server

Note: Follow the specific Zotero MCP's documentation to provide it with your Zotero API Key and User ID if required.

Step 2: Configure your Environment

Create a .env file in the root of your project (or set it in your universal shell config like ~/.zshrc).

# Point the scholarly-search capability to your local Zotero MCP node script
RESEARCH_MCP_SCHOLARLY_SEARCH_CMD="npx -y @zcaceres/zotero-mcp-server"

The exact command matches whatever you would normally type in the terminal to start the MCP server. Our Orchestrator will spawn this command as a subprocess, pipe JSON into its stdin, and read the JSON response from stdout.

Step 3: Verify the Connection

Use the doctor command to verify that the Orchestrator successfully detects your custom command hook:

python3 -m bridges.orchestrator doctor --cwd .

In the Check Details output, look for the scholarly-search provider. It should now reflect your custom npx command instead of the default Python implementation.

Reverting to Output Internet Search

If you want to cast a wider net again and search the whole internet via our default Semantic Scholar integration, simply unset the variable:

unset RESEARCH_MCP_SCHOLARLY_SEARCH_CMD