infino-ai
Keyword, semantic, hybrid and SQL retrieval over local or hosted Infino data
infino-ai is an open-source MCP server for databases. It runs locally on your machine and is started by your AI client.
- Pricing
- Open source
- Platforms
- CLI, Self-hosted, macOS
- Runs
- Locally
- Version
- 0.20.0
- Listing updated
Similar MCP servers
Install
Run in a terminal
claude mcp add mcp-server -- npx -y @infino-ai/mcp-serverAdd to claude_desktop_config.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@infino-ai/mcp-server"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@infino-ai/mcp-server"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mcp-server": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@infino-ai/mcp-server"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@infino-ai/mcp-server"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@infino-ai/mcp-server"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Configuration
INFINO_MCP_URIData to serve: a local path, an s3://|az:// bucket URI, or a hosted https://<host>/<database> endpoint (Infino Cloud). Optional — defaults to a durable per-user directory (~/.infino/mcp) so data persists across restarts, falling back to an ephemeral memory:// catalog only if that directory can't be created.INFINO_API_KEYAPI key for a hosted (https://) Infino Cloud endpoint. Required when INFINO_MCP_URI is an https:// URI; ignored for local and object-storage connections.INFINO_MCP_EMBED_PROVIDEREmbedding provider: 'local' (Hugging Face transformers.js, default) or 'openai' (any OpenAI-compatible /embeddings endpoint, including Azure OpenAI's /openai/v1 surface). Inferred as 'openai' when INFINO_MCP_EMBED_BASE_URL is set.INFINO_MCP_EMBED_BASE_URLBase URL of the OpenAI-compatible embeddings API (e.g. https://api.openai.com/v1 or https://<resource>.openai.azure.com/openai/v1). Required when the provider is 'openai'; the server POSTs to <base>/embeddings.INFINO_MCP_EMBED_API_KEYAPI key for the 'openai' provider. Sent as both Authorization: Bearer and api-key so one value works for OpenAI and Azure OpenAI.INFINO_MCP_EMBED_MODELEmbedding model. local: a Hugging Face feature-extraction model (default Xenova/all-MiniLM-L6-v2, 384-dim). openai: the model/deployment name (e.g. text-embedding-3-small, 1536-dim). Must match the model that produced the table's stored vectors.INFINO_MCP_VALIDATESet (1/true/yes) to probe the object store at startup, so bad credentials or an unreachable bucket fail then instead of on the first search.INFINO_MCP_ASKSet (1/true/yes) to offer infino_ask on an Infino Cloud account with the agent tools on. Each ask runs a model loop billed to the account.
About infino-ai
Run it with npx and it serves a data location of your choice: a local folder, an S3 or Azure bucket, or a hosted Infino Cloud database reached with an API key. With nothing configured it uses a persistent folder under your home directory, so data survives restarts, and falls back to an in-memory catalogue only if that folder cannot be created.
Semantic search embeds queries with a local model of roughly 90 MB, downloaded once and cached, so nothing leaves the machine for embedding unless you switch the provider to an OpenAI-compatible endpoint. Tables keep data and search indexes in plain Parquet files that DuckDB or pyarrow can open.
The README positions it as a retrieval layer for RAG, agent memory and search over your own files. It is also packaged as a Claude Code plugin with a usage skill and an /infino-search command, and every tool carries MCP annotations so clients can ask before destructive calls.
What it does
- Four retrieval modes
- Keyword, semantic, hybrid and SQL queries over the same data.
- Local embeddings
- Semantic search uses a local model by default, with no key needed.
- Flexible storage
- Local path, S3 or Azure bucket, or hosted Infino Cloud endpoint.
- Parquet-based tables
- Data and indexes sit in plain Parquet files other tools can read.
- Claude Code plugin
- Adds a skill and an /infino-search command in one install.
- Annotated tools
- Tools declare hints so clients can confirm destructive actions.
Good for
- Give an agent persistent searchable memory
- Search a folder of notes semantically
- Query indexed data with SQL from a chat
- Build a RAG layer over object storage
- Point an agent at a hosted Infino database
Pricing
Open source
infino-ai FAQ
What do I need to run infino-ai?
Node.js 20 or newer and an MCP client; configuration is through environment variables.
Where is data stored by default?
In ~/.infino/mcp, falling back to an ephemeral memory catalogue if that directory cannot be created.
When is an API key needed?
When INFINO_MCP_URI is an https Infino Cloud endpoint.
Does embedding send my text elsewhere?
Not by default; the local provider runs on your machine. An OpenAI-compatible provider is optional.
Is infino-ai free?
Yes. infino-ai is open source, so you can use and self-host it for free.
How do I add infino-ai to Claude Code?
Run this command in your terminal: claude mcp add mcp-server -- npx -y @infino-ai/mcp-server
What are the best alternatives to infino-ai?
The closest alternatives to infino-ai are Vectros, Opensolr MCP, Aide MCP Server, Cross-Agent Memory and SynapCores.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.