Algolia MCP Server Cursor IDE Setup 2026

How do you set up the Algolia MCP server in Cursor? Open the MCP servers page in your Algolia dashboard, pick the application and up to five indices you want an agent to query, and Algolia generates a hosted URL in the form https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp. Add that URL under mcpServers in ~/.cursor/mcp.json — no separate API key or header required, since the endpoint already carries your Search API key's permissions. Restart Cursor, confirm a green status, then ask it to search one of the indices you selected.

This page covers Algolia's Public MCP, the hosted, dashboard-configured server most people mean when they say "Algolia MCP." Algolia also documents a second, separate offering called Productivity MCP for internal workflows scoped to a signed-in user's own permissions, and two open-source reference servers you can self-host (a Go implementation at algolia/mcp and a Node.js one at algolia/mcp-node). This guide is about the hosted Public MCP path, which is what Cursor's mcp.json example in Algolia's own docs targets.

Why this is different from most MCP setups on this site

Most servers documented on this site need a token you paste into an Authorization header. Algolia's Public MCP doesn't work that way — the access control happens earlier, at the point where you create the MCP instance in the dashboard and choose which indices and Recommend models it can see. Once that URL exists, any client that can reach it inherits exactly that scope. There's no separate bearer token to rotate in Cursor's config, which also means there's no separate token to leak from mcp.json if you commit it by accident — the URL itself is the credential, so treat it with the same care you'd give an API key.

Quick reference

Prerequisites

Step 1: Create the MCP instance in the Algolia dashboard

1. Log into the Algolia dashboard and open the application you want an agent to search against.
2. Navigate to the MCP servers section of the dashboard.
3. Choose an existing MCP server URL if you've already created one, or create a new instance.
4. When creating a new instance, select up to five indices you want exposed, plus any Recommend models you want the agent able to query for recommendations.
5. Algolia generates the URL: https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp. {APP_ID} is your Algolia application ID; {UNIQ_ID} is a unique identifier tied to that specific MCP instance, not your account as a whole.

Copy the full URL. That's the only value Cursor needs.

Step 2: Add the server to Cursor

{
  "mcpServers": {
    "Algolia": {
      "url": "https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp"
    }
  }
}

Replace both placeholders with the values from Step 1. Save to ~/.cursor/mcp.json for a connection available in every project, or .cursor/mcp.json if only one repo should see this index. Restart Cursor and check Settings → Tools & Integrations → MCP Tools for a green dot next to Algolia.

There's nothing to put in an env or headers block for the hosted Public MCP path — that's the main way this config differs from almost every other server on this site. If you paste in a snippet expecting an ALGOLIA_API_KEY env var, you're looking at the self-hosted server's config, not this one; see the note at the bottom of this guide.

Step 3: Verify the connection

Ask Cursor something that requires a live query against one of the indices you selected:

Search my products index for "wireless keyboard" and list the top 5 results with their prices.

A real result set back means the URL and index selection are correct. No results at all, when you know matching records exist, points at the index not being one of the five selected when the MCP instance was created — go back to the dashboard and check, rather than assuming the query syntax is wrong.

Available tools

Algolia's Public MCP exposes a small, predictable tool surface rather than a general-purpose API client:

If you're connecting from ChatGPT rather than Cursor, Algolia's docs note that client additionally requires generic search and fetch tools alongside the Algolia-specific ones — not relevant to a Cursor setup, but worth knowing if the same MCP instance gets reused across clients.

Usage and billing

Every tool call an agent makes through Algolia's Public MCP counts as a normal Algolia API request against your account's usage — the same metering as a request from your application's search bar. There's no separate "MCP tier" that's free or unlimited. If you're wiring an agent into a heavily-used production index, treat each agent query the same way you'd treat a spike in real user traffic when thinking about your plan's request volume.

Practical workflows

Content and catalog QA:

Search the docs index for pages that mention "deprecated" and list their titles and URLs so I can review which ones need updating.

Debugging why a query returns nothing:

Search the products index for "keyboard" with no filters, then again filtered to category:electronics, and tell me if the filtered version drops results a plain search finds.

Recommendation sanity checks:

Get recommendations for product ID SKU-4821 and tell me if any of the returned items are out of stock based on their metadata.

Troubleshooting

Server never appears in Cursor. Confirm the file is .cursor/mcp.json or ~/.cursor/mcp.json, the top-level key is mcpServers, and the JSON is valid — a single syntax error drops every server defined in that file, not just this one.

Connects, but every search returns empty. The index you're querying almost certainly wasn't one of the up to five indices selected when the MCP instance was created in the dashboard. Go back to the Algolia dashboard, open that MCP instance, and check which indices it actually covers — you can't add an index to an existing instance's scope from Cursor's side.

Tool names look different from what you expected. Tool names follow algolia_search_{index_name}, using your actual index name. If your index is named prod_products, the tool is algolia_search_prod_products, not a generic algolia_search. Ask Cursor to list available tools if you're unsure what got generated.

You need write access, not just search. The Public MCP path documented here is search and recommendations only — there's no tool for writing or updating records. Index management and record writes aren't part of this offering; that's a deliberate scope limit, not a bug.

You actually wanted the self-hosted server. If a config snippet you're following references ALGOLIA_APP_ID, ALGOLIA_API_KEY, ALGOLIA_WRITE_API_KEY, or MCP_SERVER_TYPE environment variables, that's Algolia's open-source algolia/mcp (Go) or algolia/mcp-node server, run locally or self-hosted — a different setup from the dashboard-hosted Public MCP this guide covers. Both are legitimate official Algolia projects; they're just not interchangeable configs.

When not to use this

Don't point the Public MCP at an index containing anything you wouldn't want a language model reading verbatim into its context — support tickets with customer PII, unredacted internal documents, and similar. The access model here is index-level, not row-level or field-level, so if an index has sensitive records mixed in with safe ones, scope a separate index (or a different MCP instance built on a filtered replica) rather than granting broad access and hoping the agent only asks for the safe parts.

Frequently Asked Questions

What's the short version of Algolia MCP Cursor setup? In the Algolia dashboard, create an MCP instance scoped to up to five indices, copy the generated URL (https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp), and add it as a plain url entry under mcpServers in Cursor's mcp.json — no header or API key needed at the client level.

Do I need an API key in mcp.json? No, for the hosted Public MCP path. Access is controlled entirely by which indices you selected when creating the MCP instance in the dashboard, and the URL itself carries that scope.

Can I search more than five indices with one MCP server? Not with a single MCP instance — five is the limit set when you create it in the dashboard. Create a second MCP instance (a second URL, a second entry in mcp.json) if you need to cover more indices.

Can this write or update records in my Algolia index? No. The Public MCP server documented here exposes search and recommendation tools only. There's no write tool in this offering.

What's the difference between Public MCP and Productivity MCP? Public MCP, covered here, is for search and recommendation access to indices you explicitly select. Productivity MCP is a separate, internal-workflow-focused offering that inherits the signed-in user's own Algolia permissions rather than a curated index list — check Algolia's docs directly if that's the one you need, since the setup differs from what's on this page.

Related guides

Official docs cited