MCP tools

Your agents reach your archive through its MCP server at https://app.the-archivist.cc/mcp, after you approve them. This page is generated from the server’s own definitions each time the docs are built, so it always matches what agents are told.

Every tool but add_note only reads. No tool can delete anything.

What agents are told

When an agent connects, the server gives it these instructions:

This is the user's personal archive: decades of their email and notes. Use `search` for questions about what was said or written; it returns short passages with their document IDs, at most two per document. Use `get_document` to read a whole thread or note. Use `list_documents` for "everything in a period, folder or with a person". Use `people` or `get_person` for who someone is, how the user knows them and when they were last in touch; facts there cite their sources. Use `timeline` for what happened when. Use `add_note` only when the user asks you to save something. Quote the archive faithfully and say when it doesn't contain an answer.

search

Search the archive. Hybrid keyword and semantic search over the archive. Returns short passages (at most two per document) with document IDs, plus warnings.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
query	text	yes		What to look for, in words likely to appear in the text
k	whole number	no	1 to 30; default 10	How many passages (max 30)
source	one of "email", "note", "quick-note"	no
date_from	text	no		A date, YYYY-MM-DD
date_to	text	no		A date, YYYY-MM-DD
person	text	no		A person's ID, or part of a name or address
path_prefix	text	no

list_documents

List documents. Documents in a period, folder or with a person, without searching their text.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
source	one of "email", "note", "quick-note"	no
date_from	text	no		A date, YYYY-MM-DD
date_to	text	no		A date, YYYY-MM-DD
person	text	no
path_prefix	text	no
sort	one of "date", "-date", "path"	no	default "date"
limit	whole number	no	1 to 200; default 50
offset	whole number	no	at least 0; default 0

get_document

Read a document. The text of a thread or note by ID, in pages of characters. Email threads show each message under its sender and date.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
id	text	yes
offset	whole number	no	at least 0; default 0	Characters to skip
max_chars	whole number	no	1 to 100000; default 20000

people

Find people. People whose name or address matches, best match first.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
name	text	yes

get_person

A person's profile. Who someone is: relationship, summary, addresses, first and last contact, and facts (the user's own first, then by confidence) citing source document IDs, whose titles are in `documents`. Facts come in pages; `moreFacts` says how to get the next. Takes a person's ID, or a name that matches one person.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
id	text	yes		A person's ID from people, or their name
facts_offset	whole number	no	at least 0; default 0	Skip this many facts
facts_limit	whole number	no	1 to 200; default 40	How many facts (default 40, max 200)

timeline

What happened when. Events in a date range with their people, in pages (`moreEvents` says how to get the next), plus document counts per month by source.

Reads the archive. Changes nothing.

Parameter	Type	Required	Limits	Meaning
date_from	text	yes		A date, YYYY-MM-DD
date_to	text	yes		A date, YYYY-MM-DD
events_offset	whole number	no	at least 0; default 0	Skip this many events
events_limit	whole number	no	1 to 200; default 50	How many events (default 50, max 200)

add_note

Save a note. Saves a note to the archive. Use only when the user asks you to save something.

Adds to the archive.

Parameter	Type	Required	Limits	Meaning
title	text	yes	up to 500 characters
text	text	yes