Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Referna Opens Member Meeting Recaps to Personal AI Agents via MCP

Referna, a referral network for independent professionals, has launched a read-only integration that lets members bring their own live meeting recaps into a personal AI agent, supporting both the Model Context Protocol and an HTTP API. Members create an integration key, connect a compatible AI client, and can revoke access whenever they choose, with every request checked against their own participation in the meeting.

Published

Referna, a referral network for independent professionals, has introduced a read-only integration that lets members access their own Referna live meeting recaps through a personal AI agent. According to AiThority, the integration supports the Model Context Protocol, or MCP, as well as an HTTP API, and is designed so members can move that context into an AI workflow of their choosing instead of copying each recap by hand.

Referna's model places independent professionals into circles of peers who serve similar clients without competing directly. Members use those circles to build relationships, meet and make relevant introductions, and their meeting recaps capture the context of those conversations. The value of the data lies less in any single summary than in the relationships and situation it accumulates over time.

The mechanics are straightforward. After creating an integration key in Referna Settings, a member can connect a compatible AI client to list available recaps and retrieve a specific one. The returned material includes the member's personalised summary and agenda coverage, which an agent can use to help review prior conversations, prepare for a follow-up meeting, or identify points the member wants to revisit.

Control stays with the member. They decide which AI client to connect and can revoke access when they no longer want to use it, and the key itself can be replaced. Matan Protter, co-founder, CTO and head of product at Referna, said: “We believe personal AI agents will become part of how people manage their day. An AI-native network has to give those agents access to meeting recaps through MCP, with members in control.”

The boundaries are drawn explicitly. The integration is limited to the member's own recaps, and each request checks that the member participated in the meeting and was a recipient of the recap. It does not provide meeting recording files or full transcripts, and the current recap format does not promise a separate list of action items. What is being opened up, in other words, is readable context rather than raw meeting material.

MCP is spreading from model vendors and developer tools into vertical industry networks, becoming a common interface through which personal agents can reach member-owned data. When a platform built for one professional community treats MCP as its default doorway, it signals that the protocol is shifting from a technical convention into a customer-facing integration standard, and that permission design is now part of the product surface.

What to watch next is the granularity of that permission model. Read-only access has landed; the open question is whether write access follows, letting agents perform actions on a member's behalf, and how data is handled inside third-party clients after access is revoked. For professionals hoping agents can take over scheduling and meeting notes, the pace at which vertical networks open up will determine how much real work a personal agent can actually manage.

Why it matters

Connecting MCP to a vertical professional network widens the data surface personal agents can reach, and makes fine-grained, revocable permissions a competitive differentiator for similar products.

MCPAI AgentReferna
Back to realtime news

Nearby Updates

All

10/15, 14:53

Zhipu Drops GLM-5.3-BF16 on Hugging Face with Open Weights

Zhipu GLM has published GLM-5.3-BF16 to its official zai-org organization on Hugging Face, putting BF16-precision weights in a public repository alongside a linked arXiv paper. The model page already shows 40,145 downloads, with tags identifying a glm_moe_dsa architecture and support for English and Chinese conversation.

10/15, 14:49

Ant International Brings Agent Account Services to Cross-Border Trade at the Canton Fair

According to a report from Xiaguangshe, Ant International launched an agent-based account service at the Canton Fair, framing it as part of the AI transformation of cross-border trade. Cross-border trade spans many currencies, markets, and compliance steps, which makes it a natural fit for agentic automation.

10/15, 13:51

Anthropic Opens Singapore Office and Signs AI Safety Pact With Government

Anthropic has opened an office in Singapore and signed an artificial intelligence safety agreement with the country's government, according to Computer Weekly. The move pairs a physical expansion in the Asia-Pacific region with safety cooperation aimed at a national regulator, a sign that AI safety work is shifting from corporate self-regulation toward joint government arrangements.

10/15, 12:30

Figma rolls out new features that define how agents work on its platform

Figma used its first Config event in India to announce features that define how its own and third-party agents should work inside a company's design system, led by the Markdown-based Figma Specs editor. The company also moved its code layers feature into public beta and said its MCP now supports more than 20 new tools.