plexward

CONNECTORS / READ AI

Read AI

Meeting intelligence. It records and summarizes meetings, and Plexward turns that into a written note in the entity's brain plus tracked action items.

Brain layerMeeting transcripts and action items into the brain and tracker.

It is one of the highest-value connectors per minute of setup, because it closes the loop between "we agreed in the call" and "it is on the board".

What it gives Plexward

  • Transcripts and summaries for an entity's meetings, pulled by URL, by id, by "latest", or by searching for a name or date.
  • A structured brain note written into the entity's project directory, so the decision is findable months later.
  • Action items, extracted and pushed into the entity's tracker behind a confirmation gate.

Layer

brain. It is a remote HTTP MCP server declared in the entity's .mcp.json, which means entities that use it are explicit and entities that do not never see it.

{ "name": "read-ai", "layer": "local MCP", "svc": "Meetings", "account": "Acme workspace" }

Prerequisites

  • A Read AI account with the entity's meetings in it.
  • The meetings actually recorded. The connector cannot retrieve what was never captured.

Setup steps

Add the server to the brain's gitignored .mcp.json:

{
  "mcpServers": {
    "read-ai": {
      "type": "http",
      "url": "https://api.read.ai/mcp"
    }
  }
}

The first call triggers an authorization flow in a browser. That step is yours. Declare it in catalog/declarations.json and add it to the entity's routing doc.

This connector is a good illustration of why the probe matters: an HTTP MCP server in daily use can be entirely absent from a hand-written inventory, and nobody notices because it works.

Per-entity routing

## Owned Integrations
| MCP Tool | Service | Account | Notes |
|---|---|---|---|
| read-ai | Meetings | Acme workspace | Remote HTTP MCP. Meeting source of truth for this entity. |

## NOT This Company
- Meetings belonging to another entity. One Read AI account may hold meetings
  for several of them: always confirm the meeting is this entity's before
  writing anything.

One meeting source per entity. If an entity uses Read AI, it does not also use another recorder, or you will get two notes for one meeting that disagree about what was decided. Name the source in the routing doc.

Verify

cd ~/brains/acme-brain && claude mcp list | grep -i read-ai

Expected ✔ Connected. Then fetch the latest meeting and confirm it is one of this entity's.

Common failures

  • A recorder that did not join. No meeting, no transcript, and the sync reports nothing found. Check the calendar integration on the Read AI side.
  • Cross-entity meetings. A call with two of your companies in it belongs to one of them. Decide, and say so, or the action items land twice.
  • Speaker attribution in a multilingual call is imperfect. Treat the transcript as evidence, not as a verbatim record, and keep the decision summary human-reviewed.
  • Authorization expiry presents as Needs authentication in the probe.

What Plexward does with it once connected

The meeting-sync skill pulls a meeting, writes a structured markdown note into the entity's project directory, extracts the action items, and pushes them to the entity's tracker list. It never writes to a tracker without an explicit approval gate.

Because the note lands in the brain, the knowledge index picks it up, the daily brief can reference it, and next month's search finds the decision.

Every connector, one convention.

Plexward keeps the bookkeeping: which entity is supposed to have which connector, whether it is actually working right now, and what an agent is allowed to do with it.