Once you have connected an agent, it can call these seven tools. You do not call them by hand. This page is here so you know what your agent can and cannot do.

Five tools only read. Two can write, and both stop and ask you before they change anything.

search_meetings

Search across all your meetings. This is the entry point: it returns session IDs that every other tool needs.

Argument	Type	Meaning
text	string, optional	Keyword or topic to match against meeting titles and bodies. Leave it out to match every meeting, for example when you only want a date range.
participant	string, optional	Only meetings with a participant whose name contains these words. Not case sensitive, matched word by word.
after_ms	number, optional	Only meetings created at or after this time, in milliseconds since 1970.
before_ms	number, optional	Only meetings created at or before this time.
limit	number, optional	How many results. Default 20, maximum 100.

Returns total_results and a list of meetings, each with session_id, title, created_at_ms, score and participants.

get_transcript

Argument: session_id.

Returns the transcript for that meeting as text, with speakers attributed. Speaker names only appear if you have named them, which is a Pro feature. See speaker identification.

get_notes

Argument: session_id.

Returns the notes you typed yourself during the meeting, raw. Not the AI summary.

get_summary

Argument: session_id.

Returns the AI-written summary notes. There can be more than one, one per template you applied, so this returns a list under summaries.

list_action_items

Argument: session_id.

An action item is a Markdown checkbox (- [ ] something) in your memo or in an enhanced note. There is no separate to-do database. The checkbox in the note is the to-do.

Returns a list under action_items, each item carrying:

• text - the item, without the checkbox marker
• done - true or false
• note_id - which note the item lives in
• line - the line number inside that note

append_note

This one writes. Arguments: session_id, content.

Adds Markdown to the end of a meeting’s agent note. Three things make it safe:

1. It only ever writes to one machine-owned file, _agent.md, inside that meeting’s folder. Your own notes are never opened for writing, so they cannot be changed by an agent.
2. Your agent has to ask you first. The host shows a confirmation prompt, and declining cancels the write.
3. Every write is logged. See the audit log below.

mark_action_item_done

This one writes. Arguments: session_id, item_text, and optionally note_id.

Ticks a checkbox. item_text must be the exact text list_action_items gave you. Pass note_id as well when the same wording appears in more than one note; leave it out and the first note holding the item wins.

Only the two characters inside the brackets change. The rest of the line is untouched. If the item is already ticked, nothing is written at all. If no note holds that item, you get a clear error and no file changes.

The audit log

Every write, from either tool, appends one line of JSON to .mcp-audit.log at the top of your vault. Each line records the time, which tool ran, which meeting it touched, and a SHA-256 fingerprint of the target file both before and after.

The file is append-only. Old lines are never rewritten. If you ever want to know exactly what an agent changed and when, that file is the record.

cat "$HOME/Library/Application Support/ghosty-notes/.mcp-audit.log"