MCP and company-knowledge normalise
What happened
- Supabase and Composio published connector guides showing how the Model Context Protocol (MCP) links AI tools to platform data. - OpenAI emphasised 'company knowledge' for ChatGPT Business and is expanding enterprise reach via partners like Infosys. - Standards for managed knowledge layers and reusable connectors are emerging, reducing bespoke integration work for enterprise AI. (supabase.com) (techcrunch.com)
Why it matters
The plumbing for enterprise AI is getting standardized. In April, Supabase and Composio published fresh Model Context Protocol guides as OpenAI pushed “company knowledge” and signed a new distribution deal with Infosys. (supabase.com) (docs.composio.dev) (techcrunch.com) Model Context Protocol, or MCP, is a shared format for connecting an artificial intelligence tool to outside systems, the way USB gives devices one port instead of a custom cable for each accessory. Supabase’s documentation says MCP lets language models interact with Supabase projects, while Composio’s docs show developers exposing app connections through an MCP URL. (supabase.com) (docs.composio.dev) Supabase’s current guide points developers to a hosted MCP endpoint at ` lets them scope access to one project, and offers a read-only mode. Its docs also say a personal access token is no longer required for the hosted setup. (supabase.com 1) (supabase.com 2) Composio is packaging the same idea as a managed connector layer. Its docs tell developers to create a user session, use native tools or `session.mcp.url` for MCP, and connect apps on demand; its API reference lists endpoints to create, generate, update, and delete MCP servers. (docs.composio.dev 1) (docs.composio.dev 2) OpenAI has been framing the enterprise side less as “chatbots” and more as controlled access to internal data. In an October 2025 product post, the company said “company knowledge” brings context from connected apps into ChatGPT for Business, Enterprise, and Edu, with citations, admin controls, and security and privacy controls. (openai.com) (help.openai.com) That push widened on April 22, 2026, when Infosys said it would integrate OpenAI models and products including Codex into its Topaz platform. Infosys said the initial focus is software engineering, legacy modernization, and DevOps for enterprise clients in 63 countries. (infosys.com) (techcrunch.com) The common thread is less custom integration work. Supabase says MCP gives AI tools a unified way to create projects, query data, manage schema, deploy edge functions, and fetch logs, while Composio recommends its quickstart because “context management” is handled by the service. (supabase.com) (docs.composio.dev) The security model is getting standardized too, but it is not finished everywhere. Supabase’s authentication docs describe OAuth 2.1 flows, automatic discovery, dynamic client registration, and user approval for MCP clients, while its self-hosting docs say the self-hosted MCP server is not intended to be exposed to the public internet and lacks OAuth 2.1 today. (supabase.com 1) (supabase.com 2) That leaves a clearer split in the market. One layer manages access to company systems through reusable connectors, and another layer turns that access into answers and actions inside products like ChatGPT, Claude, Cursor, or internal agents. (supabase.com) (supabase.com) (openai.com) The result is that enterprise AI projects increasingly look less like one-off integrations and more like standard software hookups. The companies still differ on hosting, permissions, and tooling, but the connector layer is starting to look familiar across vendors. (supabase.com) (docs.composio.dev)
Key numbers
- (supabase.com 1) (supabase.com 2) Composio is packaging the same idea as a managed connector layer.
- (docs.composio.dev 1) (docs.composio.dev 2) OpenAI has been framing the enterprise side less as “chatbots” and more as controlled access to internal data.
- In an October 2025 product post, the company said “company knowledge” brings context from connected apps into ChatGPT for Business, Enterprise, and Edu, with citations, admin controls, and security and privacy controls.
- (openai.com) (help.openai.com) That push widened on April 22, 2026, when Infosys said it would integrate OpenAI models and products including Codex into its Topaz platform.
Quick answers
What happened in MCP and company-knowledge normalise?
Supabase and Composio published connector guides showing how the Model Context Protocol (MCP) links AI tools to platform data. OpenAI emphasised 'company knowledge' for ChatGPT Business and is expanding enterprise reach via partners like Infosys. Standards for managed knowledge layers and reusable connectors are emerging, reducing bespoke integration work for enterprise AI. (supabase.com) (techcrunch.com)
Why does MCP and company-knowledge normalise matter?
The plumbing for enterprise AI is getting standardized. In April, Supabase and Composio published fresh Model Context Protocol guides as OpenAI pushed “company knowledge” and signed a new distribution deal with Infosys. (supabase.com) (docs.composio.dev) (techcrunch.com) Model Context Protocol, or MCP, is a shared format for connecting an artificial intelligence tool to outside systems, the way USB gives devices one port instead of a custom cable for each accessory. Supabase’s documentation says MCP lets language models interact with Supabase projects, while Composio’s docs show developers exposing app connections through an MCP URL. (supabase.com) (docs.composio.dev) Supabase’s current guide points developers to a hosted MCP endpoint at lets them scope access to one project, and offers a read-only mode. Its docs also say a personal access token is no longer required for the hosted setup. (supabase.com 1) (supabase.com 2) Composio is packaging the same idea as a managed connector layer. Its docs tell developers to create a user session, use native tools or session.mcp.url` for MCP, and connect apps on demand; its API reference lists endpoints to create, generate, update, and delete MCP servers. (docs.composio.dev 1) (docs.composio.dev 2) OpenAI has been framing the enterprise side less as “chatbots” and more as controlled access to internal data. In an October 2025 product post, the company said “company knowledge” brings context from connected apps into ChatGPT for Business, Enterprise, and Edu, with citations, admin controls, and security and privacy controls. (openai.com) (help.openai.com) That push widened on April 22, 2026, when Infosys said it would integrate OpenAI models and products including Codex into its Topaz platform. Infosys said the initial focus is software engineering, legacy modernization, and DevOps for enterprise clients in 63 countries. (infosys.com) (techcrunch.com) The common thread is less custom integration work. Supabase says MCP gives AI tools a unified way to create projects, query data, manage schema, deploy edge functions, and fetch logs, while Composio recommends its quickstart because “context management” is handled by the service. (supabase.com) (docs.composio.dev) The security model is getting standardized too, but it is not finished everywhere. Supabase’s authentication docs describe OAuth 2.1 flows, automatic discovery, dynamic client registration, and user approval for MCP clients, while its self-hosting docs say the self-hosted MCP server is not intended to be exposed to the public internet and lacks OAuth 2.1 today. (supabase.com 1) (supabase.com 2) That leaves a clearer split in the market. One layer manages access to company systems through reusable connectors, and another layer turns that access into answers and actions inside products like ChatGPT, Claude, Cursor, or internal agents. (supabase.com) (supabase.com) (openai.com) The result is that enterprise AI projects increasingly look less like one-off integrations and more like standard software hookups. The companies still differ on hosting, permissions, and tooling, but the connector layer is starting to look familiar across vendors. (supabase.com) (docs.composio.dev)
