> ## Documentation Index > Fetch the complete documentation index at: https://infino.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Infino MCP server: retrieval for Claude, Cursor, and more > Expose Infino retrieval (keyword, vector, hybrid, and SQL) to any Model Context Protocol client — Claude Code, Claude Desktop, Cursor, VS Code — over a local catalog, your own bucket, or Infino Cloud. The [`@infino-ai/mcp-server`](https://www.npmjs.com/package/@infino-ai/mcp-server) package is a [Model Context Protocol](https://modelcontextprotocol.io) server for Infino. It gives an AI agent the same retrieval Infino exposes to the SDKs and the [CLI](/docs/cli) — keyword (BM25), semantic (vector), hybrid, and SQL — as tools it can call in conversation, from any MCP client (Claude Code, Claude Desktop, Cursor, VS Code, and others). The tool descriptions are written so the model picks the right retrieval mode by question shape, so you rarely name a tool yourself. It connects to the same targets as the rest of Infino: a local path, your own object-storage bucket, or a hosted [Infino Cloud](/docs/cloud/quickstart) database. Nothing extra runs — the server is the engine, in-process, behind the MCP transport. ## Install The server is launched by your MCP client over stdio; you don't run it directly in normal use. Every client config is the same shape — command `npx -y @infino-ai/mcp-server`, with configuration supplied as environment variables. ```bash theme={null} npx -y @infino-ai/mcp-server ``` ## Connect Point the server at your data with `INFINO_MCP_URI`. The scheme selects the backend: | `INFINO_MCP_URI` | Backend | | --------------------------- | ------------------------------------------------------------------------------ | | `~/.infino/mcp` (default) | a durable per-user directory, so a fresh install persists with no config | | `/path/to/data` | local disk | | `s3:///` | Amazon S3 (or S3-compatible: R2, MinIO, Backblaze B2 — set `AWS_ENDPOINT_URL`) | | `az:///` | Azure Blob | | `https:///` | [Infino Cloud](/docs/cloud/quickstart), the hosted service | For a local path or bucket, storage credentials come from the standard `AWS_*` / `AZURE_*` environment variables (omit them to use ambient cloud identity). For a hosted `https://` endpoint, authenticate with an API key in `INFINO_API_KEY` — no object-storage credentials are needed, since the platform owns the storage. ### Environment variables | Variable | Required | Default | Description | | --------------------------- | ----------------- | ------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | | `INFINO_MCP_URI` | No | `~/.infino/mcp` | Data to serve — see the table above. | | `INFINO_API_KEY` | With a hosted URI | — | API key (`inf_…`) for a hosted `https://` endpoint. Ignored for local and bucket connections. | | `INFINO_MCP_EMBED_PROVIDER` | No | `local` | Query-embedding provider: `local` (Hugging Face, no key) or `openai` (any OpenAI-compatible `/embeddings` endpoint, including Azure OpenAI). | | `INFINO_MCP_EMBED_MODEL` | No | `Xenova/all-MiniLM-L6-v2` | The embedding model — **must match the model that produced the table's stored vectors**. | The full list of embedding and validation variables is in the [infino-mcp README](https://github.com/infino-ai/infino-mcp#configuration). ## Client setup Add the server with the CLI. Use `--scope user` to make it available in every project. ```sh theme={null} claude mcp add infino \ --scope user \ -e INFINO_MCP_URI=/Users/me/.infino/memory \ -- npx -y @infino-ai/mcp-server ``` Or install the [plugin](https://github.com/infino-ai/infino-mcp#claude-code-plugin-one-step-install), which wires the server plus a how-to-use skill and an `/infino-search` command in one step: ``` /plugin marketplace add infino-ai/infino-mcp /plugin install infino@infino-ai ``` Add to the client's MCP config (`claude_desktop_config.json`, or `~/.cursor/mcp.json`), then restart: ```jsonc theme={null} { "mcpServers": { "infino": { "command": "npx", "args": ["-y", "@infino-ai/mcp-server"], "env": { "INFINO_MCP_URI": "/Users/me/.infino/memory" } } } } ``` VS Code (1.102+) reads servers from `.vscode/mcp.json`. Note the top-level key is `servers` and each entry declares `"type": "stdio"`: ```jsonc theme={null} { "servers": { "infino": { "type": "stdio", "command": "npx", "args": ["-y", "@infino-ai/mcp-server"], "env": { "INFINO_MCP_URI": "/Users/me/.infino/memory" } } } } ``` Any client that speaks MCP over stdio works — configure it to launch `npx -y @infino-ai/mcp-server` with the environment above. ## Connect to Infino Cloud To serve a hosted database, point `INFINO_MCP_URI` at the `https:///` endpoint and supply your API key. Everything else is identical to a local connection. The key's capabilities decide what the connection may do: a read-scoped key is refused every write. ```jsonc theme={null} { "command": "npx", "args": ["-y", "@infino-ai/mcp-server"], "env": { "INFINO_MCP_URI": "https://api.platform.infino.ws/my-database", "INFINO_API_KEY": "inf_…" } } ``` Get the endpoint URL (the last path segment is your database) and an API key from the [Infino Cloud console](https://platform.infino.ws); see [Authentication](/docs/cloud/authentication). Compaction and garbage collection are handled server-side on Infino Cloud, so they are not exposed as tools. ## Tools The server exposes discovery, search, and (optionally) write tools. Tool descriptions steer the model to the right one by question shape. | Tool | What it does | | ------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `infino_list_tables` / `infino_describe_table` | Discover the tables in the catalog and their columns. | | `infino_keyword_search` | BM25 full-text search — literal terms, identifiers, error codes. | | `infino_semantic_search` | Vector search by meaning; optional keyword pre-filter. | | `infino_hybrid_search` | Fused keyword + vector ranking in one pass. | | `infino_token_match` / `infino_exact_match` | Unranked keyword / exact-equality filters. | | `infino_count` | Count rows matching a keyword query, without fetching them. | | `infino_sql` | Counts, joins, aggregates, and filtering by column value; any single statement. | | `infino_create_database` / `infino_create_table` / `infino_drop_table` | Provision the database, create a table from a `{column: type}` descriptor (with `vector: true` the server adds an `embedding` column sized to its embedder), drop a table. | | `infino_add_documents` / `infino_update_documents` / `infino_delete_documents` | Writes: appended, replaced, or deleted rows. Rows without a vector are embedded from their text in one batch. | The full toolset is always available; there is no server-side write gate. Who may write is decided by the API key's capabilities on Infino Cloud and by which path or bucket you point the server at locally. Every tool carries MCP annotations (`readOnlyHint`, `destructiveHint`) so your client can ask for confirmation on its own terms. A tool call carries tens of rows; for a whole corpus use the [CLI](/docs/cli) or an SDK. Semantic and hybrid search embed queries **locally** in the server (no embedding API key by default), so `INFINO_MCP_EMBED_MODEL` must match the model that produced the table's stored vectors — this matters especially on a hosted or shared database that someone else ingested. See [Embeddings](/docs/guides/embeddings). ## See also * [CLI](/docs/cli) — the same retrieval from the terminal or a shell-native agent. * [Infino Cloud quickstart](/docs/cloud/quickstart) — provision a hosted database to serve. * [Quickstart](/docs/quickstart) — the SDK walkthrough. * [infino-mcp on GitHub](https://github.com/infino-ai/infino-mcp) — full tool arguments and configuration.