How do I add memory to Cursor AI?

Install Recallium MCP server using the setup guide. Connect it to Cursor through the Model Context Protocol (MCP) in Cursor settings. The setup takes less than 5 minutes.

What is an MCP server?

MCP (Model Context Protocol) is Anthropic's open standard for connecting AI assistants to external data sources. Recallium's platform is a production-ready MCP server specifically designed for memory management.

How does Recallium search work?

Recallium's retrieval engine reaches 98.4% hit@10 and 96.3% recall@10 on LongMemEval-S (7 September 2026 run); the full protocol, readers and judge are published at https://recallium.ai/benchmarks.

Recallium Cloud has a free tier, Cloud Free, with 500 memories a month and nothing to run; it is rolling out as early access. The community edition, offered since August 2025, formerly as MiniMe MCP, is free and self-hosted via Docker.

Which IDEs does Recallium support?

Recallium works with any tool that supports the Model Context Protocol, including Cursor, Claude Desktop, VS Code, Windsurf, JetBrains IDEs, and more.

Recallium: From Padawan to Jedi Master

Master your AI's memory in minutes. These prompts and workflows are the same on Recallium Cloud and the community edition - the ones you'll use daily, power-user workflows, five rules, and quick fixes.

9 sections~12 min readBeginner-friendly

60-second quick win

Try right now

Get value in 60 seconds. Copy-paste these prompts into your AI agent.

01START

recallium

→ Loads your full project context

02STORE

"Remember we chose Postgres because..."

→ Stores a decision

03CAPTURE

"Scan this codebase, document each component, design, and flows in my memories"

→ Stores a series of memories for later

04RECALL

"What did we decide about the database?"

→ Retrieves it instantly

Quick reference

Command cheat sheet

The prompts that cover most of daily use. Hover a row and hit copy.

Load contextrecallium

Save a decision"Remember we chose X because Y"

Find past work"Search my memories for auth flow"

Search uploaded docs"Search my documents for the API spec"

Resume work"Where did we leave off?"

Find patterns"What bugs keep recurring?"

Think through a problem"Help me think through X vs Y"

Track a follow-up"Create a task: migrate endpoints"

End of day"Summarize what we did today and save it"

Daily workflow

What you'll actually do every day

The 5 things you'll use every day. Say these to your agent in plain language.

Store Decisions & Code

You

"Remember this JWT implementation. Access tokens expire
in 15 minutes, refresh tokens in 7 days, stored in
HTTP-only cookies."

✓ Stored with type: code-snippet
   Files: src/auth/jwt.ts
   Tags: authentication, jwt, security

Recall Past Decisions

You

"What did we decide about database choice?"

We chose MongoDB for user preferences because:
- Schema flexibility for custom fields
- Nested documents for preference hierarchies
- Team has more MongoDB experience

Check File Context Before Editing

You

"I'm about to edit src/auth/jwt.ts. What should I know?"

Found 3 relevant memories:
1. JWT Refresh Implementation (2 weeks ago)
2. Bug Fix: Race Condition (1 week ago)
3. Security Enhancement (3 days ago)

⚠️ Token reuse detection is sensitive -
   test thoroughly if modifying refresh flow.

Search All Your Work

You

"Show me everything we've done on authentication"

Found 8 memories:
- 3 implementations (JWT, OAuth2, 2FA)
- 2 decisions (JWT over sessions, 15min expiry)
- 2 bug fixes (race condition, CORS)
- 1 performance improvement (45ms → 8ms)

Resume After a Break

You

"Where did we leave off?"

Last session (2 hours ago):
- Implemented rate limiting for login endpoint
- Created task: "Add integration tests for auth"
- Pending: Review PR for token rotation fix

Ready to continue?

Power user

Power-user workflows

Advanced features for deeper insights.

Pattern Detection

You

"What patterns have emerged in our API error handling?"

Analyzing 23 memories across 4 projects...

Common Patterns:
1. Centralized Error Middleware (3 projects)
2. Domain-Specific Exceptions (strong pattern)
3. Circuit Breaker for External APIs (emerging)

Recommendation: Standardize on patterns 1+2.

Bug Root Cause Analysis

You

"Why do we keep having authentication bugs?"

Quality Analysis:

Root Causes:
1. Race Conditions (3 occurrences) ⚠️
2. Token Expiry Edge Cases (2 occurrences)
3. CORS Misconfigurations (2 occurrences)

Recommendation: Add integration tests for
concurrent auth scenarios (40% of bugs).

Structured Thinking

You

"Help me think through migrating from REST to GraphQL"

🤔 Starting structured thinking...

Branch 1: Big Bang Migration
- Risk: HIGH, Timeline: 3-4 months

Branch 2: Gradual Migration (Facade)
- Risk: MEDIUM, Timeline: 6-8 months

Branch 3: New Features Only
- Risk: LOW, Timeline: Ongoing

Recommendation: Branch 2 - proves value
incrementally, team learns while productive.

Task Tracking with Context

You

"Remind me to add rate limiting. Link to our auth work."

Task created: "Add rate limiting to login"

Linked to:
- Memory #1234: JWT implementation
- Memory #1245: Auth bug fix

Full context available when you return!

One memory layer

Multi-tool & multi-project

One memory layer across all your tools and projects.

Research in Claude Desktop, Code in Cursor

Architect researches and documents in Claude Desktop, then switches to Cursor to implement.

Claude Desktop

You

"Research best practices for rate limiting APIs.
Save your findings to my memories."

Claude

✓ Stored research on rate limiting:
- Token bucket vs sliding window algorithms
- Redis-based distributed rate limiting
- Recommended: 100 req/min for auth endpoints

Cursor

You

"What did we research about rate limiting?"

Cursor

Found your research from Claude Desktop:
- Token bucket recommended for API rate limiting
- Use Redis for distributed systems
- 100 req/min for auth, 1000 req/min for general

Ready to implement based on these findings?

Import Patterns from Another Project

Reuse proven patterns across your codebase.

You

"In the payment-service project, how did we implement
error handling middleware? I want to use the same pattern here."

Found error handling pattern from payment-service:

1. Centralized ErrorMiddleware class
2. Custom exceptions: ValidationError, AuthError, NotFoundError
3. Each maps to HTTP status + error code
4. Correlation ID logging for tracing

Files: src/middleware/errorHandler.ts

Want me to adapt this pattern for your current project?

Cross-Project Insights

You

"What authentication patterns work best across all my projects?"

Analyzing 5 projects...

Strongest patterns:
1. JWT + refresh rotation (4 projects, 0 issues)
2. Redis session store (3 projects, stable)
3. RBAC middleware (2 projects, scaling well)

Avoid: Cookie-only auth caused CORS issues in 2 projects.

Unified Context Across IDEs

You

"What did I work on yesterday? I used VS Code in the morning
and Cursor in the afternoon."

Yesterday's work (both IDEs):

Morning (VS Code):
- Fixed auth middleware bug
- Added rate limiting research

Afternoon (Cursor):
- Implemented rate limiter based on research
- Created task: "Add rate limit tests"

All context unified regardless of which tool you used.

Make it a reflex

Build the habit

Recallium's value compounds over time. Build these rituals.

Start of Session

recallium

Loads rules, recent work, project context

After Finishing Work

"Save what we just did: [describe changes]"

Captures implementation while fresh

End of Day

"Summarize what we accomplished and save it"

Creates session summary for next time

After a Break

"Where did we leave off?"

Instant context restoration

Before Editing a File

"What context do we have for [filename]?"

Surfaces past decisions, warnings, bugs

Power user

Five rules for effective memory

Always search before editing a file

Your AI will find past implementations, warnings, and decisions you've forgotten.

Consolidate - save once after completing work

Don't save after every line. One comprehensive memory beats ten fragments.

Be specific - "chose X because Y" beats "chose X"

Future you needs the reasoning, not just the decision.

Include file paths when saving code work

Makes memories searchable by file - critical for the "before editing" workflow.

Save important work before moving on

If you solved something hard, save it. Your AI can't remember what you don't tell it.

Troubleshooting

Quick fixes

Tap a question to expand.

"AI isn't remembering things"

✗ Vague"Save this"

✓ Explicit"Remember this JWT implementation with refresh token rotation"

Look for the AI's confirmation: "✓ Stored with type: ..."

"Search returns nothing"

✗ Too specific"getUserByIdAndValidateToken"

✓ Broader"Search my memories for user authentication"

Semantic search finds related concepts - use natural language.

"Found wrong context"

✗ Ambiguous"Find my work"

✓ Specific"Find my authentication work in the payment-service project"

Include project name and topic for precise results.

"Too many search results"

✗ Broad"Show me everything"

✓ Filtered"Find recent authentication decisions in src/auth/"

Use filters: recent_only, memory_type, file_path.

Ready to put it to work?

Start with the 60-second quick win, then build the daily habits. Your AI gets smarter with every session.

Get early Cloud access →See Benchmarks GitHub →