specification: API Commons MCP Server specificationVersion: '0.1' provider: Algolia providerId: algolia generated: '2026-08-27' method: searched source: https://www.algolia.com/doc/guides/model-context-protocol sources: - https://www.algolia.com/doc/guides/model-context-protocol - https://www.algolia.com/doc/guides/model-context-protocol/public-mcp - https://www.algolia.com/doc/guides/model-context-protocol/productivity-mcp - https://www.algolia.com/developers/lp-mcp - https://github.com/algolia/skills/tree/main/skills/algolia-mcp description: >- Algolia ships TWO first-party, Algolia-managed remote MCP servers. Neither is a stdio package a developer runs locally - both are hosted HTTPS endpoints an MCP client POSTs to, which is the agent-reachable form. Both are read-only by design: Algolia states they "search, recommend, and (for Productivity MCP) report analytics" and that "neither one creates or modifies indices, records, or settings". Write and admin work is routed to the Algolia CLI. deployment: mode: remote endpoint: https://mcp.algolia.com/mcp auth: oauth verified: probed note: >- endpoint is the Productivity MCP server, the only tenant-independent URL Algolia publishes. The Public MCP server is also remote but its URL is per-instance and minted in the dashboard (https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp), so there is no single URL to record for it. No stdio package is published; the former github.com/algolia/mcp-node repository returns 404 as of this probe. servers: - name: Algolia Productivity MCP scope: user-scoped endpoint: https://mcp.algolia.com/mcp transport: http auth: oauth status: live verified: probed probe: method: POST body: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' http_status: 401 response: '{"status":401,"detail":"Unauthorized: Missing or invalid Authorization header","code":"UNAUTHORIZED","instance":"/mcp"}' www_authenticate: 'Bearer resource_metadata="https://mcp.algolia.com/.well-known/oauth-protected-resource", scope="public"' note: >- A correct RFC 9728 challenge - the server names its own protected-resource metadata document rather than returning a bare 401. Tool inputSchemas need an authenticated introspection; the tool NAMES below come from Algolia's own published Agent Skill, not from the live server. oauth: issuer: https://dashboard.algolia.com authorization_endpoint: https://dashboard.algolia.com/2/oauth/authorize token_endpoint: https://dashboard.algolia.com/2/oauth/token registration_endpoint: https://dashboard.algolia.com/2/oauth/register grant_types: [authorization_code, refresh_token] pkce: S256 client_auth: none scopes: [public] discovery: - https://mcp.algolia.com/.well-known/oauth-authorization-server - https://mcp.algolia.com/.well-known/oauth-protected-resource note: >- Dynamic client registration is open, PKCE is required, and the client is public (no secret). Algolia's docs say explicitly "You don't need to provide an OAuth client ID or client secret. If your MCP client asks for these fields, leave them empty." access_model: >- Enabled once per Algolia account, then each user signs in with their own Algolia login; that user's Algolia permissions determine which applications and indices the agent can reach. install: claude_code: claude mcp add --transport http algolia https://mcp.algolia.com/mcp codex: codex mcp add algolia --url https://mcp.algolia.com/mcp generic: 'Add a remote/custom MCP connector with URL https://mcp.algolia.com/mcp and authenticate via OAuth.' clients_documented: [ChatGPT, Claude, Claude Code, Codex, Cursor, VS Code, Gemini CLI, OpenAI Playground] - name: Algolia Public MCP scope: application-scoped endpoint: 'https://{APP_ID}.algolia.net/mcp/1/{UNIQ_ID}/mcp' transport: http auth: none status: live verified: searched note: >- Per-instance endpoint created from the Algolia dashboard (Generative AI > MCP servers). Up to five instances per application, each exposing up to five indices plus selected Recommend models. Algolia's docs state "The MCP server uses your Search API key permissions to control data access. Your MCP client doesn't require separate authentication." Not probed: the URL is templated on a tenant's own APP_ID and a per-instance UNIQ_ID, so there is no anonymous instance to call, and guessing one would be fabrication. intended_use: >- Exposing curated indices and Recommend models to external or customer-facing AI assistants (ChatGPT, Claude, Vertex AI, Perplexity, Amazon Q, Microsoft Copilot, Salesforce Agentforce). limits: instances_per_application: 5 indices_per_instance: 5 pricing: Usage counts toward the existing Algolia plan; Algolia states there is no separate MCP fee. read_only: true read_only_source: https://www.algolia.com/doc/guides/model-context-protocol tools: method: searched source: https://raw.githubusercontent.com/algolia/skills/main/skills/algolia-mcp/SKILL.md note: >- Tool names are quoted verbatim from Algolia's own published Agent Skill (saved to skills/_provider/algolia-mcp.md). Input schemas were NOT read - the live tools/list is OAuth-gated. See mcp/algolia-tool-crosswalk.yml for the binding of each tool to its backing OpenAPI operation. search: - algolia_search_index - algolia_search_list_indices - algolia_search_for_facet_values analytics: - algolia_analytics_top_searches - algolia_analytics_searches_no_results - algolia_analytics_no_results_rate - algolia_analytics_click_positions - algolia_analytics_no_click_rate - algolia_analytics_top_searches_without_clicks - algolia_analytics_number_of_searches - algolia_analytics_top_search_results - algolia_analytics_number_of_users - algolia_analytics_top_filters - algolia_analytics_top_filters_no_results - algolia_analytics_top_countries recommendations: - tool: algolia_recommendations models: [bought-together, related-products, trending-items, trending-facets, looking-similar] related_agent_surfaces: - name: Agent Studio type: Algolia-hosted agent runtime (not MCP) openapi: openapi/algolia-agent-studio-api-openapi.yml operations: 42 docs: https://www.algolia.com/doc/guides/algolia-ai/agent-studio note: >- Distinct from the MCP servers: Agent Studio is Algolia's own agent platform, and it can itself CONSUME third-party MCP tools (https://www.algolia.com/doc/guides/algolia-ai/agent-studio/how-to/tools/mcp-tools). Algolia is therefore both an MCP server and an MCP client.