RemembrallMCP

Whole-codebase knowledge for AI coding agents

The Challenge

Every time an AI coding agent starts a task, it greps through files, spawns sub-agents, and reads dozens of files just to understand what calls what. A single "find all callers of this function" question can cost 5,000-20,000 tokens and take 20+ tool calls. On large codebases, agents hit context limits, miss cross-module dependencies, or give incomplete answers because they never found all the references.

The Solution

RemembrallMCP pre-indexes your codebase into a field-aware code graph using tree-sitter. Functions, classes, methods, fields, imports, and field references become queryable in under 1ms. When an agent asks "what breaks if I change this function?" or "who reads the amount field?", it gets the answer in one tool call - not after 20 greps. It also stores decisions, patterns, and context as persistent memory that survives between sessions. Benchmarks on a real codebase show 95% fewer tool calls and 98% fewer tokens compared to standard agent exploration.

Key Features

Field-aware code graph across 9 languages - Python, TypeScript, JavaScript, Rust, Go, Java, Kotlin, Ruby, C#

Field-level references - who reads self.amount, this.balance, or s.Address

Blast radius analysis - "what breaks if I change this?" answered in under 1ms

Symbol lookup across the entire project with file and line number

Persistent memory with hybrid semantic + full-text search using RRF fusion

GitHub PR ingestion to bootstrap memory from project history

Markdown docs ingestion to populate knowledge from README, ADRs, and architecture docs

In-process embeddings via fastembed - no external API calls

Docker Compose one-command setup with Postgres and pgvector

Incremental indexing - only reparses changed files

AI Capabilities

How AI Powers RemembrallMCP

95% Fewer Tool Calls

Benchmarked against Click v8.1.7: agents without RemembrallMCP needed 112 tool calls across 5 coding tasks. With RemembrallMCP, the same tasks took 5 calls. That is 95% fewer calls and 98% fewer tokens spent on exploration.

Field-Aware Graph, Not Just Memory

Most MCP memory servers store and retrieve text. RemembrallMCP also parses your source code into a queryable graph with functions, classes, methods, fields, imports, call chains, and field references. Agents can check blast radius before making changes, trace call chains, and find who reads a given field - all in a single tool call.

Works With Any MCP Client

Standard MCP protocol over stdio. Works with Claude Code, Cursor, and any MCP-compatible AI agent. Ten tools cover memory (store, recall, update, delete), code graph (index, impact, lookup, tour), and ingestion (GitHub PRs, markdown docs).

Field-Level References Across 9 Languages

Goes beyond call relationships to capture data fields and who reads them. Ask "who references the amount field?" and get the methods that read self.amount, this.balance, or s.Address - the question ORM-heavy codebases live or die by. Covers Python, TypeScript, JavaScript, Rust, Go, Java, Kotlin, Ruby, and C#.

Tech Stack

RustMCP ServerPostgreSQLpgvectortree-sitter

Visit RemembrallMCP→

Need something like this built?

RoboticForce builds production AI systems, MCP servers, and APIs. We ship working software, not slide decks.

Start a project