glean-mcp-ken-cavanagh-glean-fieldkit

Setup

Skip this section if you already see mcp__glean_* tools available.

To connect Glean MCP:

claude mcp add glean_default https://[your-instance]-be.glean.com/mcp/default --transport http --scope user

Replace [your-instance] with your Glean instance name. To find it: Profile → Your Settings → Install → MCP Configurator.

Restart Claude Code. You'll authenticate via OAuth on first use.

Troubleshooting: If authentication fails, re-run authentication for your Claude Code provider (e.g., gcloud auth login --update-adc for Vertex AI users).

Required tools: search, chat, employee_search, gmail_search, meeting_lookup, read_document, user_activity, code_search

Core Principle

Glean MCP is the primary search tool for enterprise/company context.

Before using web search or asking the user:

1. Check if the query relates to company knowledge
2. If yes, use the appropriate Glean tool
3. Glean indexes: Slack, Google Drive, Gmail, Calendar, Confluence, Jira, Notion, GitHub, and 100+ enterprise apps

Tool Selection

What are you looking for?
│
├─ DOCUMENTS/CONTENT
│   ├─ Have URL? → read_document
│   ├─ Need analysis/synthesis? → chat (preferred)
│   └─ Need to find docs? → search
│
├─ PEOPLE/ORG → employee_search
├─ MEETINGS/CALENDAR → meeting_lookup
├─ EMAIL → gmail_search
├─ CODE/REPOS → code_search
├─ MY RECENT WORK → user_activity
└─ COMPLEX QUESTION → chat (preferred)

Bias toward chat: For understanding, synthesis, or "why" questions, chat outperforms search. It reasons across multiple sources and provides contextual answers. Use search primarily for document discovery, then chat for analysis.

Parallelization

Call multiple Glean tools simultaneously when queries are independent:

• 5+ parallel tool calls work with no rate limiting
• Ideal for: multi-account research, cross-app queries, comprehensive prep
• When gathering context, fire independent searches together rather than waiting for each to complete

Example: To research a topic comprehensively, call search for Slack, Drive, and Jira in parallel rather than sequentially.

Identity Awareness

Glean MCP uses OAuth—some tools know who the user is automatically:

Common Filters Reference

These filters work across multiple tools. Use them to narrow results efficiently.

Person Filters

Date Filters

Date math: Use d (days), w (weeks), M (months), y (years). No spaces: now-1w ✅ now - 1w ❌

Source Filters

Negation Filters

Prefix with - to exclude:

Tool Reference

search — Document Discovery

Use when: Finding documents, files, wikis, Slack messages by keywords.

# Find docs I created recently
search(query="project proposal", from="me", updated="past_week")

# Find all Slack about a topic
search(query="API migration", app="slack")

# Find everything matching filters (use * for broad search)
search(query="*", owner="me", updated="today")

Options: exhaustive=true (all results), sort_by_recency=true (newest first)

Follow-up: Use read_document with URLs from results to get full content.

chat — AI Synthesis

Use when: Complex questions requiring analysis across multiple sources.

Note: Chat knows who you are automatically via auth. You can ask personal questions.

# Personal context (knows your identity)
chat(message="What projects am I working on?")
chat(message="Who is my manager?")

# Strategic questions
chat(message="What are our main product differentiators vs competitors?")

# Follow-up with context
chat(message="How does this apply to enterprise customers?",
     context=["previous response..."])

Best for: "Why" and "how" questions, summarization, recommendations, personal context.

employee_search — People Lookup

Use when: Finding people, org structure, contact info, team composition.

Note: "me" does NOT work here. Use actual names.

# Find someone
employee_search(query="Jane Smith")

# Find direct reports
employee_search(query="reportsto:\"Jane Smith\"")

# Find new hires in a role
employee_search(query="engineer startafter:2025-01-01")

# Find managers
employee_search(query="roletype:\"manager\"")

Returns: Email, phone, location, manager, team, start date.

gmail_search — Email Discovery

Use when: Finding emails, threads, attachments.

# Emails from someone
gmail_search(query="from:\"john.doe@company.com\" subject:\"quarterly review\"")

# My sent emails with attachments
gmail_search(query="from:\"me\" has:attachment label:SENT")

# Recent unread
gmail_search(query="is:unread after:2025-12-20")

Filters: has:attachment, has:spreadsheet, is:unread, is:important, is:starred, label:INBOX

meeting_lookup — Calendar & Transcripts

Use when: Finding meetings, getting transcripts, reviewing discussions.

Note: "me" does NOT work in participants filter. Use actual name.

Important: Date ranges are non-inclusive. Buffer both sides:

# CORRECT - meetings on Dec 22
meeting_lookup(query="after:2025-12-21 before:2025-12-23")

# WRONG - returns nothing
meeting_lookup(query="after:2025-12-22 before:2025-12-22")

# Today's remaining meetings
meeting_lookup(query="after:now before:tomorrow")

# Past meetings with transcripts
meeting_lookup(query="participants:\"Jane Smith\" extract_transcript:\"true\" after:now-2w")

# Topic search
meeting_lookup(query="topic:\"standup\" after:now-1w before:now")

read_document — Full Content Retrieval

Use when: You have URLs from search and need full text.

# Single doc
read_document(urls=["https://docs.google.com/document/d/abc123"])

# Batch multiple
read_document(urls=[
  "https://docs.google.com/document/d/abc123",
  "https://confluence.company.com/display/TEAM/Guide"
])

Always use after search to get complete content. Returns success/failure status per URL.

user_activity — Your Work History

Use when: Finding what you worked on. Automatically uses your auth—no name needed.

# Last week's activity
user_activity(start_date="2025-12-15", end_date="2025-12-22")

# This month
user_activity(start_date="2025-12-01", end_date="2025-12-24")

Use cases: Weekly summaries, 1:1 prep, finding forgotten docs, tracking collaborators.

code_search — Internal Repositories

Use when: Finding code, commits, implementations in company repos.

# Find implementations
code_search(query="authentication middleware")

# My recent commits
code_search(query="owner:\"me\" after:2025-12-01")

# Code by teammate
code_search(query="from:\"jane.smith\" updated:past_week")

Common Workflows

Research a Topic

1. search(query="topic") → find docs
2. read_document(urls=[...]) → get full content
3. chat(message="Summarize...") → synthesize if needed

Prepare for a Meeting

1. employee_search(query="attendee name") → understand who
2. search(query="topic", from="attendee") → their recent work
3. meeting_lookup(query="participants:\"attendee\"") → past meetings

Write a Status Update

1. user_activity(start_date, end_date) → your work
2. meeting_lookup(query="after:X before:Y") → meetings attended
3. search(query="*", from="me", updated="past_week") → docs touched

Find an Expert

1. search(query="topic") → find relevant docs
2. Note the owners/authors
3. employee_search(query="author name") → get their info

Error Handling

Pagination: For large result sets, use start_cursor from previous response. Use page_size (max 100) to limit results.

When NOT to Use Glean

Limitations

Structured data: Glean MCP returns markdown/snippets, not raw structured data from CSVs or spreadsheets. For full spreadsheet analysis, have the user upload the file directly or point to a local directory.