Documentation
Everything you need to get your team up and running with Recall.
Quick Start
1. Sign up and create a team
Go to recall.team and sign in with GitHub. Create a team and select your repositories.
2. Install the MCP server
Copy your API token from the dashboard and add Recall to your Claude Code configuration:
{
"mcpServers": {
"recall": {
"command": "npx",
"args": ["-y", "@anthropic/recall-mcp"],
"env": {
"RECALL_API_TOKEN": "your-token-here"
}
}
}
}
3. Start coding
That's it! Recall will automatically load team context at the start of each session and save summaries when you're done.
How It Works
Memory Files
Recall creates three types of memory files in your repository's .recall/ folder:
- context.md (~1.5-3K tokens) - Team brain, loads every session
- history.md (~30K tokens) - Encyclopedia for onboarding and deep dives
- sessions/ - Individual session records (~1.5K each)
The Flow
context.md automatically via MCPrecall_save_session to save a summary.recall/ to share with your teamHotwords
Use these keywords to load different levels of context:
| Keyword | What Loads | Token Budget |
|---|
| (default) | context.md | ~1.5-3K |
| "remember" | history.md | ~30K |
|---|
| "ultraremember" | sessions/ | Full transcripts |
|---|
MCP Tools
Available Tools
Recall provides these MCP tools to Claude Code:
recall_get_context
Loads team context (context.md). Called automatically at session start.
recall_get_history
Loads detailed session history (history.md). Use when you need to remember past decisions.
recall_get_transcripts
Loads full session transcripts from sessions/ folder. Use sparingly - high token usage.
recall_save_session
Saves a summary of the current session. Call this before ending your work.
recall_save_session({
summary: "Implemented user authentication with GitHub OAuth",
decisions: [
{ what: "Used NextAuth for auth", why: "First-party Next.js integration" }
],
filesChanged: ["src/app/api/auth/[...nextauth]/route.ts"],
nextSteps: "Add session persistence"
})
recall_log_decision
Quick way to log a single decision during coding.
recall_auth
Authenticate with your Recall API token. Usually not needed if token is in env.
File Structure
.recall/ Folder
After initialization, your repo will have:
.recall/
├── context.md # Team brain (encrypted)
├── history.md # Session history (encrypted)
├── sessions/ # Individual sessions
│ └── 2025-01/
│ └── ray/
│ └── 15-1430.md
├── events/
│ └── events.jsonl
├── manifest.json
└── README.md
Encryption
All memory files are encrypted with your team's key. Only authenticated team members can decrypt them.
The encryption key is stored on Recall servers. Your data lives in YOUR GitHub repo - we just manage the keys.
Team Management
Adding Team Members
Roles
- Owner: Full control, can delete team
- Admin: Can invite/remove members
- Member: Can use Recall, can't manage team
Billing
Each team member needs a seat. Add/remove seats anytime from the dashboard.
Seats are prorated - you only pay for what you use.
Best Practices
Saving Sessions
Save at natural breakpoints:
- End of a feature
- Before a long break
- When you've made important decisions
Use specific summaries:
- Bad: "Worked on the app"
- Good: "Added pagination to user list with cursor-based approach"
Team Context
Keep context.md focused:
- Current project state
- Recent decisions
- Active blockers
Move historical info to history.md:
- Past decisions and reasoning
- Lessons learned
- Failed approaches
Git Workflow
Commit .recall/ with your code:
git add .recall && git commit -m "Update team context"
Use meaningful commit messages for recall changes.