mise

Guide for content fetching — web URLs, Google Drive, Gmail — using the mise-en-space MCP.

Iron Law: Files are artifacts. Emails are meaning.

A document tells you what was decided. The email thread tells you why, who pushed back, and what concerns remain.

The 3-Verb Model

mise-en-space exposes 3 tools:

Tool	Purpose	Output
`search`	Find files/emails	Path to deposited JSON + counts
`fetch`	Extract content	Path to deposit folder with content.md, comments.md
`create`	Make new docs	File ID + URL

Filesystem-first: Results go to mise-fetch/ in the specified directory. Read what you need.

Critical: Always Pass base_path

MCP servers run as separate processes. Without base_path, deposits land in the MCP server's directory — not yours.

python

# ALWAYS pass base_path when calling via MCP
search("Q4 planning", base_path="/Users/modha/Repos/my-project")
fetch("1abc...", base_path="/Users/modha/Repos/my-project")

This puts mise-fetch/ next to your project where you can read it. Without it, files disappear into the server's install directory.

CLI users: mise search / mise fetch from the command line use your shell's cwd automatically — no base_path needed.

The Exploration Loop

Don't just search→fetch→read. Follow the loop:

code

1. Search Drive → find file
2. Check email_context → if present, the file came from an email
3. Search Gmail filename:X → find the email thread
4. Read email → discover new terms, people, context
5. Search again with new terms → repeat

This loop discovers meaning (in communications) behind artifacts (files).

See references/exploration-loop.md for details.

After Every Fetch

mise automatically deposits alongside your content:

•comments.md — open comments (if any)
•manifest.json — metadata including open_comment_count

Always check for comments.md. The real discussion often lives in comments — disagreements, questions, suggestions that didn't make it into the final text.

bash

# What's in my deposit?
ls mise-fetch/doc--strawman-framework--1abc/
# content.md  comments.md  manifest.json

# Don't skip the comments!

Gmail Search: Use Operators

Don't: Throw keywords at search (keyword soup)

python

# BAD
search("Elizabeth Kiernan Lantern data privacy contracts")

Do: Use Gmail operators for precision

python

# GOOD
search("from:elizabeth@privacylawunlocked.com after:2025/12/01", sources=["gmail"])
search("filename:strawman from:legal@thinkbox.tv", sources=["gmail"])

See references/gmail-operators.md for full reference.

Web Content Fetching

fetch also handles web URLs — any http:// or https:// that isn't a Google service:

python

# Blog posts, documentation, articles
fetch("https://simonwillison.net/2024/Dec/19/one-shot-python-tools/")

# GitHub raw files (markdown, code, config)
fetch("https://raw.githubusercontent.com/fastapi/fastapi/master/pyproject.toml")

# API responses (JSON)
fetch("https://api.github.com/repos/anthropics/anthropic-cookbook")

What it handles:

•HTML pages — Extracts main content via trafilatura, removes boilerplate
•Code blocks — Preserves language hints (python, typescript, etc.)
•Raw text — Markdown, JSON, Python, TOML passed through directly with code fences
•JS-rendered pages — Falls back to browser (requires webctl start)

Deposits to: mise-fetch/web--{title}--{hash}/content.md

Limitations:

•CAPTCHA-protected sites (Wikipedia, some news) will fail
•Heavy JS apps may need browser fallback
•Auth-required pages need cookies (future feature)

Filtering Large Results

When search returns drive_count: 20, gmail_count: 15:

Don't: Read the entire 35-result JSON file

Do: Filter with jq first

bash

# Preview top 5
jq '.drive_results[:5] | .[] | {name, id}' mise-fetch/search--*.json

# Filter by name
jq '.drive_results[] | select(.name | test("framework"; "i"))' mise-fetch/search--*.json

See references/filtering-results.md for patterns.

Quick Reference

Search (parallel across sources)

python

search("Q4 planning", base_path="/path/to/project")              # Both Drive + Gmail
search("budget 2026", sources=["drive"], base_path="/path/to/project")  # Drive only
search("from:boss@co.com", sources=["gmail"], base_path="/path/to/project")  # Gmail only

Fetch (auto-detects type)

python

fetch("1abc...", base_path="/path/to/project")                    # Drive file ID
fetch("https://docs.google.com/...", base_path="/path/to/project")  # Drive URL
fetch("18f3a4b5c6d7e8f9", base_path="/path/to/project")          # Gmail thread ID
fetch("https://example.com/blog/post", base_path="/path/to/project")  # Web URL

Create (markdown → Google Doc)

python

create("# Meeting Notes\n\n- Item 1", title="Team Sync")
create(content, title="Report", folder_id="1xyz...")

Anti-Patterns

Pattern	Problem	Fix
Keyword soup in Gmail	Poor precision	Use `from:`, `filename:`, `after:` operators
Read full search JSON	Token waste	Filter with jq first
Skip comments.md	Miss real discussion	Always check after fetch
Ignore email_context	Miss source thread	Fetch the linked email
Stop after first search	Shallow research	Follow the exploration loop

Large Fetch Deposits

For big email threads (32k+ tokens) or long docs, filter before full Read:

bash

# Preview first 50 lines
head -50 mise-fetch/gmail--re-lantern--abc123/content.md

# Grep for specific topic
grep -A5 "controllership" mise-fetch/gmail--*/content.md

# Count messages in thread
grep -c "^## Message" mise-fetch/gmail--*/content.md

When to Use

•Fetching web content (cleaner than curl/WebFetch — extracts article, removes boilerplate)
•Research tasks involving multiple Drive/Gmail sources
•Finding context around a document (who sent it, what was discussed)
•Assembling background for meetings or strategic documents
•Any task where you need to trace communication threads

When NOT to Use

•Single-file operations (just use fetch directly)
•You already know the patterns
•Simple lookups that don't need exploration

Success Criteria

This skill works when:

•You use Gmail operators instead of keyword soup
•You check comments.md after every doc fetch
•You follow email_context hints to find source emails
•You filter large results before reading
•You follow the exploration loop for research tasks