Search links

Search links let AI chatbots (like ChatGPT, Claude, or your own) search your knowledge base via public URL.

When to use search links

Scenario	Use search links
Large knowledge base (100+ docs)	Perfect
AI needs to find specific answers	Perfect
Dynamic Q&A over your docs	Perfect
Small set of always-needed docs	Use Instruction sets instead

How it works

1. Create a search link for your workspace
2. Share the URL with AI or integrate in your app
3. AI searches your documents and gets relevant chunks

Your documents -> Synjar indexes -> AI searches via URL -> Gets answers

Creating a search link

1. Go to your workspace
2. Click the Search Links tab
3. Click Create Search Link
4. Configure:
   • Name (optional): For your reference, e.g., "Support KB"
   • Scope: All documents or specific tags
   • Expiration: Never or specific date

Using with ChatGPT or Claude

You have two options:

Option 1: MCP Integration (Recommended - Automatic)

MCP (Model Context Protocol) allows ChatGPT and Claude to search your knowledge base automatically - no manual link pasting needed!

Setup time: Less than 2 minutes

Read the MCP Setup Guide

What you get:

• Automatic searching (no link pasting)
• Works with ChatGPT Developer Mode and Claude Custom Connectors
• Same privacy settings as direct search

---

Option 2: Manual Link Pasting (Fallback)

If you don't have access to MCP (Developer Mode / Custom Connectors), you can use the manual workaround:

Why the assistant asks you to paste the link

This is a safety restriction in LLM browsing tools: they cannot open URLs they generated themselves. The workaround is simple: paste the exact link into the chat when the assistant asks for it. If multiple searches are needed, paste all links the assistant provides.

When to use this:

• You don't have ChatGPT Developer Mode or Claude Custom Connectors
• You prefer a simpler setup (no configuration needed)

See full instructions in MCP Connectors guide

Using in your application

For developers building custom integrations:

const response = await fetch(
  'https://api.synjar.com/public/YOUR_TOKEN/search?q=refund+policy'
);
const { results } = await response.json();

// results contains matching document chunks with scores
results.forEach(result => {
  console.log(result.title, result.score);
  console.log(result.content);
});

See the API Reference for full documentation.

Limiting access with tags

To restrict which documents are searchable:

1. Tag your documents (e.g., public-facing, internal)
2. When creating search link, select Only specific tags
3. Choose allowed tags (e.g., only public-facing)

Now only documents with those tags will be included in search results.

Security

What they CAN do	What they CANNOT do
Search your documents	Edit or delete documents
Read matched chunks	See full document content
Filter by allowed tags	Access other workspaces

Best practices

• Use tag scoping for public-facing links
• Set expiration for time-limited access
• Revoke links when no longer needed

Search links vs instruction sets

Feature	Search links	Instruction sets
How it works	AI searches on-demand	AI loads full content
Document limit	Unlimited	~20 documents
Use case	Q&A, large KB	Brand voice, procedures
Content freshness	Always current	Updated on rebuild

tip

You can use both together! Use instruction sets for core instructions and search links for reference material.