Algolia MCP Server
OfficialQuery and manage Algolia search indices from an agent: search, settings, analytics, A/B tests, and Recommend.
Add to your client
Copy the config for your MCP client and paste it into its config file.
git clone git@github.com:algolia/mcp.git && cd mcp/cmd/mcp && go buildPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"algolia-mcp-server": {
"command": "/path/to/the/repo/cmd/mcp/mcp",
"args": [],
"env": {
"ALGOLIA_APP_ID": "<your-app-id>",
"ALGOLIA_INDEX_NAME": "<your-index-name>",
"ALGOLIA_API_KEY": "<your-api-key>",
"ALGOLIA_WRITE_API_KEY": "<your-admin-api-key>"
}
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- Go toolchain installed (go.dev/doc/install) to build the binary
- Algolia application ID, index name, and API key; admin key only if you want write operations
- An MCP client such as Claude Desktop (README builds on the official MCP quickstart)
About Algolia MCP Server
Algolia's MCP server lives under the official algolia GitHub org and gives agents structured access to the Algolia platform, not just query execution. Sample prompts from the README show the range: filtered product searches, inserting JSON objects into an index, updating searchable attributes, and tuning ranking — the last three requiring the write key.
Tool enablement is config-driven. MCP_ENABLED_TOOLS accepts a comma-separated list from: abtesting, analytics, collections, monitoring, querysuggestions, recommend, search, search_read, search_write, usage; empty means everything. The split between search_read and search_write is the practical security lever — you can run a read-only search agent without ever configuring the admin key.
It is a Go binary you compile yourself (cd cmd/mcp && go build) and reference by absolute path in the MCP config. Transport defaults to stdio; setting MCP_SERVER_TYPE=sse with MCP_SSE_PORT (default 8080) runs it as an SSE server instead. The README also documents running it under mcphost with local Ollama models.
Know the caveats before adopting: Algolia calls it a prototype to explore, explicitly not officially supported, and reserves the right to break it. For production-minded setups Algolia points to its hosted MCP (announced in the README banner) and the algolia/mcp-node project. Treat this repo as the fast-moving reference implementation.
Tools & capabilities (10)
searchEnables all search operations, both read and write.
search_readRead-only search: list indices, get settings, run queries, get objects.
search_writeWrite operations: clear, copy, delete, move indices; set settings; insert/delete objects.
collectionsWork with Algolia collections.
analyticsAccess Algolia Analytics data.
abtestingManage and inspect A/B tests.
querysuggestionsWork with Query Suggestions configuration.
recommendAccess Algolia Recommend.
monitoringMonitor cluster/index status.
usageInspect Algolia usage data.
What this server can do
Algolia MCP Server provides tools for these capabilities — tap one to see every MCP server that does the same:
When to use it
- Ask natural-language questions over an index ('all Nike products under $100')
- Let an agent tune relevance: update searchable attributes or ranking from a conversation
- Insert or clean up records in an index during data workflows
- Review analytics and A/B test results without opening the dashboard
Quick setup
- 1git clone git@github.com:algolia/mcp.git
- 2cd mcp/cmd/mcp && go build, then note the absolute binary path (pwd)
- 3Add the binary path as command in claude_desktop_config.json with ALGOLIA_APP_ID, ALGOLIA_INDEX_NAME, ALGOLIA_API_KEY env vars
- 4Optionally add ALGOLIA_WRITE_API_KEY (admin key) for writes and set MCP_ENABLED_TOOLS to restrict tool groups
- 5Restart Claude Desktop and confirm the algolia tools appear
Security notes
With only the search API key it is effectively read-mostly; adding ALGOLIA_WRITE_API_KEY hands the agent your admin key, which can delete indices and rewrite settings — enable search_write only when needed. Algolia explicitly says this repo is experimental, unsupported, not covered by SLA, and may break or disappear; MCP usage counts against your Algolia plan.
Algolia MCP Server FAQ
Is the Algolia MCP server production-safe?
Algolia says no: the repo is experimental, not officially supported, outside the SLA, and may change or be removed at any time. It is fine for exploration; for production leanings use Algolia's hosted MCP documented on algolia.com, and never give the admin key to an agent you don't supervise.
Can I make it read-only?
Yes, two ways: omit ALGOLIA_WRITE_API_KEY so write tools have no credentials, and/or set MCP_ENABLED_TOOLS to search_read plus whichever analytics-style groups you want. Write operations only work with the admin key configured.
Why does it need Go instead of npx?
The server is written in Go and distributed as source; you compile a native binary with go build and point your MCP config at its absolute path. Algolia maintains a separate Node.js variant at algolia/mcp-node if you prefer the npm ecosystem.
Alternatives to Algolia MCP Server
Compare all alternatives →
Official MCP reference server that fetches a URL and returns its content as clean Markdown, with chunking.
Official Firecrawl MCP server — scrape, crawl, map, search, and structured extraction for any LLM client.
Exa's neural web search and crawling MCP server — runs locally via npx or as a hosted remote endpoint.
Compare Algolia MCP Server with: