Obot is an open source control plane for enterprise AI, built by Obot AI, Inc.
It gives security and platform teams one place to govern the AI agents, chat clients and Model Context Protocol (MCP) servers their organisation uses, so that staff can connect AI tools to company data without every team wiring up its own access.
At A Glance
| Vendor | Obot AI, Inc. |
|---|---|
| Category | Agentic AI identity, MCP gateway |
| Identities covered | Human users and groups, AI agents and MCP clients, MCP servers, upstream service credentials |
| Deployment | Managed cloud, or self-hosted on Docker or Kubernetes |
| Agentic AI | MCP access control, credential isolation, Nanobot agent framework |
| Funding | Seed stage, venture backed |
| Market classification | Agentic AI Identity |
What Obot Does
As companies adopt AI assistants and agents, each one tends to need its own connection to email, documents, data and internal systems. That quickly produces unmanaged tools, scattered OAuth grants and little record of what agents actually did. Obot puts a governed layer in the middle. Administrators approve which MCP servers and Skills are available, users sign in through the organisation’s identity provider, and every tool call and LLM request passes through Obot, where it is checked and logged. The whole platform is open source and can run in the customer’s own cloud or on premises, so organisations keep control of the data their agents touch.
Products
AI Control Plane
The AI Control Plane is the governance view over agent activity. Security teams set policy on what agents may do, find tools that were connected outside approved channels, and keep full logs of every tool call and LLM request, which can be exported to the organisation’s own storage.
Obot MCP Gateway
The gateway solves the problem of giving many AI clients safe access to many MCP servers. It hosts servers behind one endpoint with “One OAuth setup across every MCP server you connect”, puts SSO and OIDC in front of each connection and applies content filters to MCP traffic. Developers can reach approved servers from any IDE or application without manual setup. The project is MIT licensed, and its sister project Nanobot is an open source framework for building agents on MCP and MCP-UI.
MCP Catalog
The catalogue is the approved list. Administrators review MCP servers and Skills before publishing them to users, and the published set includes servers for Word, Google Sheets, OneDrive, Outlook and Kensho by S&P Global.
How It Works
Obot sits between AI clients and user devices on one side and hosted services and external providers on the other. It can be deployed on any Kubernetes cluster, in any cloud or on premises, or used as a managed cloud service.
| Component | What it does |
|---|---|
| LLM Gateway | Sits between chat clients and model providers to monitor and control model traffic. |
| Hosted and remote servers | Obot runs each hosted MCP server in its own Docker container or Kubernetes workload, while remote servers stay on external infrastructure. |
| Data store | A PostgreSQL database holds configuration, metadata and audit data, with optional application-level encryption for selected fields. |
| Token handling | Clients receive Obot tokens, and Obot validates the access token on every MCP request before forwarding it. |
Identities And Systems Covered
| Human users and groups | Sign in through the organisation’s identity provider, with access based on ownership and user and group policies. |
|---|---|
| MCP clients and agents | OAuth with PKCE for public clients; approving a client does not by itself grant access to any server. |
| Upstream credentials | Held by Obot for each connected service, kept separate from client tokens. |
| Systems | Microsoft Word, Outlook (through the Microsoft Graph API), OneDrive, Google Sheets and Kensho, through catalogue MCP servers. |
Agentic AI Capabilities
Scoped Tool Access
Administrators build virtual MCP endpoints (vMCPs) that expose selected tools from one or more servers. A shared vMCP gives each user or group only the tools they have been granted, while a personal vMCP is usable only by its owner.
Credential Isolation
Every request carries two separate authorisations: the user’s right to reach the MCP server, and the server’s own credentials for the system behind it. Because upstream credentials never leave Obot, an agent or client cannot reuse them elsewhere.
Integrations
| Area | Integrations |
|---|---|
| Identity providers | GitHub, Google, Okta, other OIDC-compatible providers |
| AI clients | Claude Code |
Company Background
| Legal entity | Obot AI, Inc. develops and maintains the open source projects. |
|---|---|
| Leadership | Sheng Liang, CEO. |
| Funding | A $35 million seed round announced on 23 September 2025, co-led by Mayfield Fund and Nexus Venture Partners. |
Note: This overview is compiled from Obot’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.