Open Source - MIT License

Give your agent the whole codebase, not a few pages.

A field-aware code graph plus persistent memory. Your AI coding agent stops grepping and re-deriving how everything fits - it already holds the whole codebase.

95%Fewer tool calls

98%Token savings

<1msGraph queries

Get Started View on GitHub

Works with Claude Code, Cursor, and any MCP-compatible client

THE PROBLEM

Every AI agent reads a few pages out of the book. Every single session.

Stateless by default

> "What calls UserService?" Agent: Searching... reading file 1/8... Agent: Reading file 2/8... Agent: Spawning sub-agent... Agent: Reading file 3/8... Agent: Reading file 4/8... ... Agent: UserService is called by AuthController, OrderHandler, and NotificationService Cost: ~5,000 tokens | 15 tool calls | 12s

With RemembrallMCP

> "What calls UserService?" remembrall_impact: upstream UserService Result: AuthController.authenticate() OrderHandler.process() NotificationService.send() Cost: ~200 tokens | 1 tool call | <1ms

Before and after: chaotic exploration vs structured graph queries

Your agents re-explore your codebase from scratch every session - a few pages out of the book. RemembrallMCP indexes the whole thing once and answers instantly.

TWO CAPABILITIES

A field-aware code graph, plus memory that survives

Field-Aware Code Graph

Pre-indexes your codebase with tree-sitter. Functions, classes, methods, fields, imports, and field references - queryable in sub-millisecond.

✓ 9 languages: Python, Java, JavaScript, Rust, Go, Ruby, TypeScript, Kotlin, C#
✓ Fields and references: who reads self.amount, this.balance, s.Address?
✓ Blast radius analysis - what breaks if I change this?
✓ Symbol lookup: find where any function, class, or field is defined
✓ Cross-file import resolution and incremental re-indexing

Persistent Memory

Stores decisions, patterns, and context that survive between sessions. Hybrid semantic + full-text search finds what matters.

✓ 9 memory types: Decision, Pattern, ErrorPattern, Preference, Architecture, and more
✓ Hybrid search: pgvector semantic + Postgres full-text + RRF fusion
✓ Cold start solved: ingest from GitHub PRs or markdown docs
✓ Scoped by organization, team, or project
✓ Importance scoring and access tracking

BENCHMARKS

Real numbers. Real codebase. No cherry-picking.

Tested against pallets/click v8.1.7 (594 symbols) with 5 real coding tasks.

Metric	Without RemembrallMCP	With RemembrallMCP	Improvement
Total tool calls (5 tasks)	112	5	-95.5%
Estimated tokens	~56,000	~1,000	-98.2%
Avg calls per question	22.4	1.0	-95.5%

MEMORY QUALITY

31/31queries passed (100%)

0.917Recall@5

14msP95 latency

0.908MRR

10 MCP TOOLS

Everything an agent needs. Nothing it doesn't.

Memory

remembrall_store

Store knowledge with vector embeddings

remembrall_recall

Hybrid semantic + full-text search

remembrall_update

Update content, tags, importance

remembrall_delete

Remove memories by UUID

Code Graph

remembrall_index

Parse project, build field-aware code graph

remembrall_impact

Blast radius on functions, classes, methods, and fields

remembrall_lookup_symbol

Find where functions, classes, methods, or fields are defined

remembrall_tour

Generate guided reading-order tour of indexed codebase

Ingestion

remembrall_ingest_github

Import merged PR descriptions as memories

remembrall_ingest_docs

Scan markdown files, split by headers, auto-classify

BUILT WITH

Pure Rust. Zero Python. No external APIs.

RustPostgreSQL 16pgvectortree-sitterfastembed (ONNX)MCP Protocol

All embeddings generated in-process with ONNX Runtime. No API calls, no Python dependencies, no token costs for indexing. Your code never leaves your machine.

Index your codebase in 2 minutes.

Quickstart with Docker Compose

git clone https://github.com/roboticforce/remembrallmcp.git
cd remembrallmcp
docker compose up -d

Add to your MCP client config

{
  "mcpServers": {
    "remembrall": {
      "command": "remembrall"
    }
  }
}

View on GitHub View Case Study