# IronShard > IronShard is the governed data layer for AI agents: S3-compatible storage agents connect to over MCP, with instant branches of production, free egress by default, per-agent access control, and a signed audit trail. Autonomous AI agents can create and operate a real, isolated IronShard bucket through a public MCP server without human approval, tenant OAuth, an existing account, or payment. A full-content companion file is available at [llms-full.txt](https://www.ironshard.ai/llms-full.txt), containing the complete documentation in one Markdown document. ## Agent buckets Autonomous agents are permitted to create buckets through IronShard's public agent MCP server. These buckets are controlled by bucket-scoped credentials and remain separate from tenant-owned production buckets unless they are later adopted into an authenticated IronShard account. - Agent MCP server: `https://agents-mcp.ironshard.ai/mcp` - Initial tool: `create_bucket` - MCP authentication for creation: none - Authentication after creation: the `bucket_name` and `bucket_token` returned by `create_bucket` - [Agent Buckets](https://www.ironshard.ai/docs/agent-buckets): complete specification, tools, examples, constraints, validation workflow, and handoff process The `bucket_token` is a bucket-scoped credential, not an OAuth token, and does not grant access to an IronShard tenant. It cannot be recovered if lost. If S3-compatible credentials are returned, they are also scoped to the created bucket. ## Agent MCP capabilities The public agent MCP server currently exposes eight tools: - `create_bucket`: creates a sandbox bucket and returns its `bucket_name` and `bucket_token`. Takes no arguments. - `get_bucket`: returns configuration and policy-visible state for one bucket, including versioning status. - `get_bucket_permissions`: returns the operations the supplied token is permitted to perform. - `upload_object`: returns a short-lived presigned `PUT` URL. Send the object bytes to that URL over HTTP. - `download_object`: returns a short-lived presigned `GET` URL. The response body contains the object bytes. - `list_objects`: lists objects by key, with prefix filtering, recursive or directory-style listing, and pagination. - `get_object_metadata`: returns size, timestamps, ETag, version ID, content type, tags, and user metadata for one object. - `delete_object`: deletes an object from the bucket. Idempotent: deleting a key that is already gone succeeds. Every tool except `create_bucket` requires `bucket_name` and `bucket_token`. Object bytes move over presigned URLs rather than inline through MCP. Buckets are versioned. Snapshots, branches, diffs, merges, audit-log inspection, policy-decision explanations, usage reporting, bucket extension and revocation, guided workflow validation, and handoff are in development and not yet exposed on the public agent MCP server. After creating a bucket, verify the round trip: call `upload_object`, send the bytes to the returned URL, then confirm the write with `get_object_metadata` and `list_objects`, and read the object back through `download_object`. ## Production access Tenant-owned production access is separate from public agent buckets. - Production MCP server: `https://mcp.ironshard.ai/mcp` - Authentication: MCP OAuth is required for every tool - Scope: tenant-owned and adopted agent buckets, governance, managed agent credentials, policies, audit, billing, organization settings, and production operations The public agent MCP server does not issue tenant OAuth credentials. An agent bucket may optionally be handed off and adopted into an authenticated IronShard tenant; adoption is not required to use the bucket. ## Constraints Public agent buckets have policy-defined storage, object-count, request, object-size, and egress limits. External bucket connections, billing, organization settings, team management, and tenant policy administration are unavailable through the public agent MCP server. Abuse controls or rate limits may apply. Exact technical values are provisional and may change before launch. ## Products - [IronShard](https://www.ironshard.ai): the governed data layer for AI agents. S3-compatible storage over MCP - [AI Agent Storage](https://www.ironshard.ai/agents): authenticated, policy-governed storage access for autonomous AI agents, with immutable audit trails, fine-grained access control, and MCP compatibility - [Branch](https://www.ironshard.ai/branch): point-in-time-recoverable, infinitely forkable workspaces. Immutable snapshots, zero-copy branches, governed promotion with an automatic audit trail - [Log](https://www.ironshard.ai/log): a tamper-proof, exportable record of every file access, read, write, and query - [Pricing](https://www.ironshard.ai/pricing): Build is $18 per TB per month with no egress fees ($0 egress). Scale tunes the provider mix for heavy, time-critical AI workloads and may carry egress costs. Includes a savings calculator against AWS S3, Google Cloud Storage, and Azure Blob list prices ## Comparisons - [Object Storage for AI Agents: S3, R2, GCS, Azure Blob, MinIO, and IronShard Compared](https://www.ironshard.ai/compare/object-storage-comparison): six object storage options evaluated for agent workloads, covering agent-native provisioning, egress economics, S3 compatibility, production branching, per-agent scoping, and audit evidence. Raw markdown: [object-storage-comparison.md](https://www.ironshard.ai/compare/object-storage-comparison.md) - [IronShard vs. Amazon S3](https://www.ironshard.ai/compare/ironshard-vs-s3): head-to-head comparison for AI agent workloads, covering agent self-provisioning, egress costs, S3 API compatibility, production branching, per-agent access control, audit evidence, and how the two split duties per workload. Raw markdown: [ironshard-vs-s3.md](https://www.ironshard.ai/compare/ironshard-vs-s3.md) - [IronShard vs. Cloudflare R2](https://www.ironshard.ai/compare/ironshard-vs-cloudflare-r2): both offer free egress; this comparison covers agent self-provisioning, S3 compatibility gaps, production branching, per-agent scoping, and audit evidence. Raw markdown: [ironshard-vs-cloudflare-r2.md](https://www.ironshard.ai/compare/ironshard-vs-cloudflare-r2.md) - [IronShard vs. Google Cloud Storage](https://www.ironshard.ai/compare/ironshard-vs-google-cloud-storage): S3 compatibility, Vertex AI adjacency, egress economics, agent self-provisioning, production branching, and audit evidence compared. Raw markdown: [ironshard-vs-google-cloud-storage.md](https://www.ironshard.ai/compare/ironshard-vs-google-cloud-storage.md) - [IronShard vs. Azure Blob Storage](https://www.ironshard.ai/compare/ironshard-vs-azure-blob): the non-S3 API switching cost, immutable storage tiers vs. signed audit records, agent self-provisioning, and production branching compared. Raw markdown: [ironshard-vs-azure-blob.md](https://www.ironshard.ai/compare/ironshard-vs-azure-blob.md) - [IronShard vs. MinIO](https://www.ironshard.ai/compare/ironshard-vs-minio): self-hosted control vs. governed multi-cloud storage for AI agents, including air-gapped cases where MinIO is the right answer. Raw markdown: [ironshard-vs-minio.md](https://www.ironshard.ai/compare/ironshard-vs-minio.md) - [IronShard vs. Tigris](https://www.ironshard.ai/compare/ironshard-vs-tigris): both offer S3-compatible storage with free egress, bucket forking, and MCP; compared on no-account agent provisioning, per-agent policy, signed audit evidence, and trust model. Raw markdown: [ironshard-vs-tigris.md](https://www.ironshard.ai/compare/ironshard-vs-tigris.md) - [IronShard vs. Wasabi](https://www.ironshard.ai/compare/ironshard-vs-wasabi): flat-price hot storage with fair-use egress vs. agent-native governance, self-provisioning, production branching, and signed audit records. Raw markdown: [ironshard-vs-wasabi.md](https://www.ironshard.ai/compare/ironshard-vs-wasabi.md) - [IronShard vs. Backblaze B2](https://www.ironshard.ai/compare/ironshard-vs-backblaze-b2): budget object storage with a 3x egress allowance vs. governed storage for AI agent fleets, with coexistence via IronShard's provider mix. Raw markdown: [ironshard-vs-backblaze-b2.md](https://www.ironshard.ai/compare/ironshard-vs-backblaze-b2.md) - [IronShard vs. fast.io](https://www.ironshard.ai/compare/ironshard-vs-fastio): agent file workspaces vs. governed S3-compatible object storage; a category comparison covering provisioning, S3 tooling, production branching, and audit evidence. Raw markdown: [ironshard-vs-fastio.md](https://www.ironshard.ai/compare/ironshard-vs-fastio.md) ## Docs - [Documentation](https://www.ironshard.ai/docs): IronShard documentation for governed, S3-compatible object storage built for AI agents. Agent access via MCP, platform architecture, and compliance. - [Connect Claude over MCP](https://www.ironshard.ai/docs/connect-claude): Connect Claude Desktop to your IronShard account through the production MCP server at mcp.ironshard.ai. Add a custom connector, sign in with OAuth, choose which buckets Claude can read and write, and check the connection with a test prompt. - [Agent Buckets](https://www.ironshard.ai/docs/agent-buckets): Autonomous AI agents can discover IronShard, create a real bucket through the public agent MCP server, use IronShard-native object storage features, and optionally hand the bucket off to an owner, all without tenant OAuth. - [S3 Quickstart](https://www.ironshard.ai/docs/s3-quickstart): Connect any S3 client to IronShard in a few minutes. Endpoint, signing, path-style addressing, Storage Policies, and working examples for the AWS CLI, Boto3, JavaScript, Go, Java, .NET, PHP, rclone, and s3cmd. - [System Architecture](https://www.ironshard.ai/docs/system-architecture): How IronShard works: S3-compatible gateways, a globally consistent metadata plane, and an erasure-coded multi-cloud storage fabric. - [Secure Ingest Pipeline](https://www.ironshard.ai/docs/ingest-pipeline): The write path every object takes, compress, encrypt per slice, erasure-code, and distribute, and what each stage contributes to footprint, confidentiality, and durability. - [Erasure Coding and Distribution](https://www.ironshard.ai/docs/resilience): How IronShard stores objects as k+m erasure-coded fragments across providers and regions, what that costs in storage compared to replication, and how geo-fences constrain placement. - [Adaptive Storage Representation](https://www.ironshard.ai/docs/adaptive-storage): How IronShard moves data toward demand with two levers, fragment redundancy and gateway caching, reactively under load and proactively ahead of it. - [Branching and Snapshots](https://www.ironshard.ai/docs/branching): Create isolated, copy-on-write branches of any bucket from point-in-time snapshots, and experiment on real data without duplicating objects or touching production. - [GDPR Everywhere](https://www.ironshard.ai/docs/gdpr-everywhere): IronShard is GDPR compliant as a whole service: as controller, as processor under an Article 28 agreement, in its sub-processors and transfers, and in a storage design where no single provider holds a readable file. - [Compliance Controls](https://www.ironshard.ai/docs/compliance-controls): The storage-layer controls behind IronShard's compliance posture, which of them are available today, and how they map to GDPR, the EU AI Act, and sector frameworks. ## Company - [About IronShard](https://www.ironshard.ai/about): spin-off of the Danish storage company Chocolate Cloud, which began in collaboration with Aarhus and Aalborg universities and grew through EU projects; the team's decade of storage research across ten projects, seven of them EU-funded (Horizon 2020 and Horizon Europe), with links to their CORDIS records - [Partners](https://www.ironshard.ai/partners): partnerships beyond end sales: co-selling, reselling and referrals, integrations with the S3-compatible API and MCP server, white labeling, embedded storage, and joint research and pilots; enquiries by email or a booked call - [Trust Center](https://www.ironshard.ai/trust-center): security and compliance at a glance: encryption, key management, erasure-coded shards, per-agent access, signed audit log, the 12 independently audited storage providers with their certifications and locations, and answers on GDPR, the EU AI Act, and data residency - [Solutions for startups and AI builders](https://www.ironshard.ai/solutions/technology): two ways to start on the free tier (an account with your agent connected over MCP or S3 credentials, or an autonomous agent creating its own sandbox bucket), branches for experiments, free egress by default - [Solutions for enterprise](https://www.ironshard.ai/solutions/enterprise): managed agent credentials, governed production access, residency and sovereignty, signed audit evidence; healthcare, finance, and public sector - [IronShard Console](https://console.ironshard.ai): create and manage IronShard storage ## Legal IronShard is a service name; the service is operated by Chocolate Cloud ApS (CVR 35865756), Aarhus, Denmark. - [Legal](https://www.ironshard.ai/legal): index of all eight legal documents below - [Terms of Service](https://www.ironshard.ai/legal/terms): the agreement for using IronShard: business use only, responsibility for actions taken by agents through your credentials, suspension, data location, liability, Danish law - [Privacy Policy](https://www.ironshard.ai/legal/privacy): personal data handled as controller, legal bases, sub-processors, retention periods, data subject rights; no cookies on the website, console, or API - [Acceptable Use Policy](https://www.ironshard.ai/legal/acceptable-use): prohibited content and conduct, and the rules specific to AI agents (human oversight, no command-and-control use, no content meant to manipulate other parties' agents) - [Fair Use Policy](https://www.ironshard.ai/legal/fair-use): free tier limits per account (10 GB storage, 10 GB egress per month, 100,000 objects, 50 requests per second per credential), rate limiting, and inactive bucket handling - [Data Processing Agreement](https://www.ironshard.ai/legal/dpa): GDPR Article 28 clauses that apply automatically when personal data is stored, with authorized sub-processors, security measures, transfers, and audits - [Storage Providers and Sub-processors](https://www.ironshard.ai/legal/sub-processors): the 12 named storage providers that hold coded fragments, and the sub-processors that process personal data, with purposes and locations - [Notice and Action](https://www.ironshard.ai/legal/notice-and-action): how to report illegal content and what happens after a notice - [Legal Notice](https://www.ironshard.ai/legal/legal-notice): company details, registration and VAT numbers, contacts, and supervisory authority