Local MCP for AI agents — Xtify | Every tool, one memory
AI agents · Local MCP

Your meeting memory, wired to your agents.

Xtify runs a Model Context Protocol server on your Mac. The AI agents you already use can recall every decision, action item, and risk — and none of it ever leaves your machine.

01 · Yes, we support local MCP

A full MCP server ships inside the app — and it runs entirely on your Mac.

Every meeting Xtify captures becomes structured memory — decisions, action items, risks, follow-ups, transcripts. Xtify exposes that memory through a Model Context Protocol server that starts with the app, listens only on localhost, and reads straight from the index on your disk.

There is no cloud endpoint to enable, no memory copied to our servers, and no path in for a third-party cloud agent. The server is on-device by design — the same local-first posture as the rest of Xtify.

On-device server
Binds to 127.0.0.1 — never the open internet
Read-only by default
Agents query memory; they can't alter it
Standard protocol
Any MCP-compatible client connects, no plugin
02 · Made for AI agents to call

Point an agent at it, and your whole meeting history becomes callable context.

Because Xtify speaks standard MCP, any agent that supports the protocol can connect in one config entry — no custom SDK, no scraping, no exporting. Ask a question in plain language and the agent decides which Xtify tool to call, pulls exactly the memory it needs, and answers with real context from your meetings.

Claude Desktop
Cursor
Cline · Zed
Your own scripts

Anything that implements an MCP client — desktop assistants, IDE agents, or a few lines of your own code — works the same way.

How it fits together

Everything stays inside one boundary — your Mac.

Your Mac
Local AI agents
Claude Desktop
Cursor / Cline
Custom scripts
Xtify MCP server
127.0.0.1 · stdio
Meeting memory
Indexed on your disk
Cloud agents have no route in — the server never leaves localhost.
What the server exposes

A focused toolset over the four extraction primitives.

Every tool maps to how Xtify already structures a meeting — decisions, action items, risks, follow-ups — plus search and raw transcripts. All read-only.

search_memory

Semantic search across every meeting. Returns ranked snippets with the meeting, speaker, and timestamp they came from.

list_meetings

Enumerate meetings, filterable by date range, platform, or participant. Returns titles, metadata, and stable IDs.

get_meeting

The full structured summary for one meeting — decisions, actions, risks, follow-ups, and speaker breakdown, by ID.

list_action_items

Open and closed action items with owner, due date, and source meeting. Filter by owner, status, or overdue.

list_decisions

Every decision on record across meetings, with who made it and the context it was made in.

list_risks

Risks and blockers with severity, who raised them, and the meeting of origin. Filter by high / medium / low.

find_mentions

Every place a person, project, or topic was mentioned — across all meetings — with the surrounding transcript.

get_transcript

The full diarized transcript for a meeting, with speaker labels and timestamps — when an agent needs the raw source.

list_followups

Open follow-ups threaded across meetings, so an agent can see what's still waiting on a reply or a next step.

Tools are also surfaced as MCP resources — xtify://meetings/{id} and xtify://memory/index — for clients that browse resources directly.

Connect in a minute

One config entry, then ask.

~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "xtify": {
      "command": "xtify",
      "args": ["mcp", "--stdio"]
    }
  }
}
1
Enable the server

Turn on Local MCP in Xtify's settings. The server starts with the app and binds to localhost only.

2
Add the config entry

Drop the snippet into your agent's MCP config and restart it. Xtify appears as an available tool source.

3
Just ask

Ask your agent about your meetings in plain language — it calls the right Xtify tool and answers with real context.

In practice

What an agent does with it.

You

"What did I commit to for Project Atlas, and is anything overdue?"

Agent calls list_action_items filtered by owner + overdue → returns two tasks, one past due Monday.
You

"Summarize every risk raised about the data-sharing agreement this quarter."

Agent calls find_mentions then list_risks → groups them by meeting and severity.
You

"Draft a follow-up email to the client covering what we decided last week."

Agent calls list_meetings then list_decisions → writes the email grounded in the actual record.
Why local

Local MCP vs. cloud MCP.

Xtify — local MCP
  • Memory index lives on your disk — nothing is copied to a vendor's cloud.
  • Only agents on your machine can reach the server.
  • Works offline — no network round-trip to answer a query.
  • No third party sits between your agent and your memory.
Typical cloud MCP
  • Transcripts and memory stored on the vendor's servers.
  • A hosted endpoint any authorized cloud agent can query.
  • Every query is a network round-trip to a third party.
  • Your memory depends on the vendor staying online.
Availability

Local MCP is on Pro Max and Ultra.

Bring your own agents on Pro Max. Ultra adds separated memory per client or workspace, so each engagement exposes its own scope through the server.

See plans

Give your agents a memory that never leaves your Mac.

Install Xtify, turn on Local MCP, and connect the agent you already use.