Going deeper

Connectors

Folio exposes your transcripts, tasks, and memory to MCP-aware tools, locally. Access is read-only, runs over stdio, and never touches the cloud.

What is MCP#

The Model Context Protocol is an open standard that lets an application expose context and tools to an AI client through a small, well-defined interface. A client connects to a server, lists the tools it offers, and calls them on demand. The protocol does not require a network. A server can run as a local process and speak to the client over standard input and output.

Folio ships a local MCP server named folio-mcp. It gives any MCP-aware tool read-only access to your transcripts, tasks, and memories over stdio. There is no cloud and no proxy. The server runs on your machine, reads from your vault, and answers the client directly. The one exception is create_task, which writes a new task back into your vault.

Setup#

Enabling connectors takes two steps. Turn the server on inside Folio, then point your MCP client at it.

  1. 1

    Enable connectors in Folio

    Open Settings then Connectors and enable the local MCP server. Folio runs folio-mcp on your machine and serves it over stdio.

  2. 2

    Register folio-mcp with your client

    Add the server to your MCP client configuration. The client launches folio-mcp as a subprocess and talks to it over standard input and output. A minimal configuration registers a single server named folio with the command folio-mcp and an empty argument list.

    mcp.json
    {
      "mcpServers": {
        "folio": {
          "command": "folio-mcp",
          "args": []
        }
      }
    }

Compatible tools#

Any tool that speaks the Model Context Protocol can connect to folio-mcp. That includes the clients people most often reach for.

  • Claude Desktop
  • Cursor
  • Claude Code
  • Any MCP client

The configuration shape varies slightly between clients, but the server entry is the same. Register folio with the command folio-mcp and no arguments.

Available tools#

The server exposes a focused set of tools over MCP. Every tool reads from your local vault. Only create_task writes.

search_memoryread
Searches your stored memories and returns the entries that match a query.
recent_meetingsread
Lists your most recent meetings, newest first.
get_transcriptread
Returns the full transcript for a single meeting.
notes_by_date_rangeread
Returns the notes that fall within a start and end date.
notes_by_folderread
Returns the notes stored under a given folder in your vault.
notes_by_personread
Returns the notes that involve a specific attendee.
quote_segmentread
Returns an exact quoted segment from a transcript with its speaker attribution.
find_decisionread
Searches transcripts for a decision and returns where it was made.
list_tasksread
Lists the tasks held in your vault.
create_taskwrite
Creates a new task and writes it back into your vault. This is the one write-style tool the server exposes.

Privacy and scope#

Local and read-only by design

Access runs over stdio and is read-only, with create_task as the only exception. The server is scoped to your local vault. Nothing is sent to a third party, and no cloud or proxy sits between your client and your notes. For the full picture of what stays on your machine, see the privacy documentation.