# Tigris — The Storage Cloud Purpose Built for AI > Tigris is globally distributed, S3-compatible object storage with zero egress fees. A single global endpoint, low-latency access from anywhere, and predictable costs — purpose built for AI training and inference workloads. ## What Tigris is Tigris is an S3-compatible object storage service that replaces AWS S3, Google Cloud Storage, and Cloudflare R2. Data is globally distributed across multiple regions automatically, so there is one endpoint (`https://t3.storage.dev`) and no per-region bucket juggling. Tigris fulfills over 90% of the S3 API. - **Globally distributed**: writes are durable across regions; reads are served from the closest region. - **Zero egress fees**: no bandwidth charges for reading your data out. - **S3-compatible**: use any AWS SDK, `aws` CLI, boto3, etc. without rewriting code. - **Predictable pricing**: simple usage-based tiers with a generous free tier. ## Get started ```bash # Install the Tigris CLI brew install tigrisdata/tap/tigris # macOS / Linux (Homebrew) npm install -g @tigrisdata/cli # Cross-platform curl -fsSL https://github.com/tigrisdata/cli/releases/latest/download/install.sh | sh # Native # Log in tigris login ``` Or configure the AWS SDK directly: ```bash export AWS_ENDPOINT_URL=https://t3.storage.dev export AWS_ACCESS_KEY_ID= export AWS_SECRET_ACCESS_KEY= export AWS_REGION=auto ``` ## Tigris MCP server An official hosted MCP server is available at `https://mcp.storage.dev/mcp` (HTTP transport). ```bash # Claude Code claude mcp add --scope user --transport http tigris https://mcp.storage.dev/mcp ``` Install instructions for Cursor, VS Code, Codex, ChatGPT, Claude Desktop, and Goose are at . ## Key resources - **Docs**: - **Get started quickstart**: - **CLI**: - **SDKs**: - **Buckets**: - **Objects**: - **IAM**: - **Migration from S3/GCS/R2**: - **llms.txt**: - **Pricing**: ## Common use cases - **AI training and inference data**: put datasets, checkpoints, and model artifacts in one globally accessible bucket and stream them to any GPU region without egress charges. - **Low-latency object serving**: static assets, user uploads, and generated media served from the nearest region automatically. - **Multi-cloud / workload portability**: one S3-compatible endpoint reachable from AWS, GCP, Azure, or on-prem without setting up cross-cloud replication. - **S3 migration**: drop-in replacement that honors the S3 API surface most apps actually use. ## Support and community - **Status**: - **Discord**: - **Blog**: - **GitHub**: