wehub-resource-sync
26382a7ac6
8.4 KiB
lean-ctx vs Aider repo-map
Last updated: May 2026 | Aider pioneered PageRank-based repo maps for AI coding. lean-ctx brings the same concept to every MCP-compatible agent.
Overview
| lean-ctx | Aider repo-map | |
|---|---|---|
| Approach | MCP-available context layer with PageRank repo-map | Built-in feature of Aider CLI |
| GitHub Stars | 2,600+ (lean-ctx) | 43,000+ (Aider) |
| Language | Rust | Python |
| Availability | MCP server (works with 28 agents) | Locked to Aider CLI |
| PageRank | Session-aware personalized PageRank | Personalized PageRank |
| Scope | 72+ MCP tools (repo-map is one) | Repo-map + AI coding assistant |
The Core Difference
Aider is a complete AI coding assistant with a built-in repo-map feature. The repo-map uses personalized PageRank to identify the most relevant files and symbols for the current conversation, presenting them as compact elided code views. It's proven technology — Aider consistently scores well on SWE-bench.
lean-ctx implements the same PageRank repo-map concept via ctx_repomap, but makes it available as an MCP tool that works with any MCP-compatible agent. It also adds session-aware personalization (recent files and task context influence rankings) and integrates with 67 other tools for a complete context engineering workflow.
The key distinction: Aider's repo-map is locked to Aider. lean-ctx's repo-map works with Cursor, Claude Code, Codex, Windsurf, Gemini, and 23 other agents.
Feature Comparison
| Feature | lean-ctx ctx_repomap | Aider repo-map |
|---|---|---|
| PageRank algorithm | Personalized power iteration | Personalized PageRank (networkx) |
| Session-aware ranking | Recent files boosted, task context weighting | Chat files boosted |
| Token budget control | max_tokens parameter (default 1024) |
--map-tokens (default 1k) |
| Binary search fitting | Yes | Yes |
| Tree-sitter parsing | 26 languages | 40+ languages |
| Symbol extraction | Functions, classes, traits, structs | Functions, classes, methods |
| Edge weighting | Proper casing +8, private x0.1, active session x50 | Frequency-based logarithmic |
| Caching | mtime-based invalidation | Persistent cache |
| Enhanced dependency maps | Via property graph | --use-enhanced-map (import-based) |
| MCP available | Yes (works with 28 agents) | No (Aider CLI only) |
| Embedding-based search | Hybrid BM25 + dense vector | Via Aider's codebase search |
| Shell compression | 95+ patterns | No |
| Session memory | Knowledge graph + temporal facts | Chat history |
| Call graph | Multi-hop BFS | No |
| Impact analysis | ctx_impact (6 actions) | No |
| Observability | Token tracking dashboard | No |
The PageRank Algorithm
Both tools use the same core idea, inspired by Google's PageRank:
- Build a graph: files are nodes, symbol definitions and references create edges
- Compute PageRank: rank files by their graph centrality (how "connected" they are)
- Personalize: boost files relevant to the current context
- Budget-fit: binary search to select top-ranked symbols within a token limit
Aider's Implementation
Source files → tree-sitter → definitions + references
↓
Graph (files as nodes, refs as edges)
↓
PageRank (personalized by chat files)
↓
Binary search → token budget fit
↓
Elided code view (scope-aware)
Aider also supports --use-enhanced-map which uses import statements to create a dependency estimator, reducing false edges from symbol name collisions.
lean-ctx's Implementation
Source files → tree-sitter → definitions + references
↓
Property graph (SQLite, 8 node types, 14 edge types)
↓
PageRank (personalized by session state)
↓
Binary search → token budget fit
↓
Compressed signatures (lean-ctx format)
lean-ctx uses its existing property graph (which also powers call graphs, impact analysis, and search ranking) instead of a separate networkx graph. The personalization vector draws from:
- Active session files: files read or edited in the current session get a boost
- Task context: if the agent has an active task, related files rank higher
- Knowledge graph: previously learned architectural relationships influence ranking
MCP Availability: The Key Advantage
Aider's repo-map is arguably the most effective codebase orientation tool for AI agents. But it's only available inside Aider's CLI — you can't use it in Cursor, Claude Code, Windsurf, or any other tool.
lean-ctx makes the same capability available as an MCP tool:
# From any MCP-compatible agent (Cursor, Claude Code, Codex, ...)
# The agent calls ctx_repomap automatically when it needs codebase orientation
# Or from the CLI
lean-ctx repomap ./my-project --max-tokens 2048
This means you get PageRank-based codebase orientation regardless of which AI coding tool you use.
Beyond Repo-Map: The Full Stack
Aider is a complete AI coding assistant — repo-map is one feature among many (inline editing, git integration, voice coding, etc.).
lean-ctx is a context engineering layer — repo-map is one tool among 68+. The difference is that lean-ctx doesn't try to be the AI coding assistant. It enhances whatever assistant you already use:
| lean-ctx Feature | Complements repo-map by... |
|---|---|
| ctx_read (10 modes) | Compressing the files that repo-map identifies as important |
| ctx_search (hybrid) | Finding specific code when repo-map gives the overview |
| ctx_callgraph | Tracing execution paths through repo-map's ranked symbols |
| ctx_impact | Understanding blast radius of changes to top-ranked files |
| Session memory | Remembering which parts of the repo-map were explored |
| Shell compression | Compressing build/test output after making changes |
Language Support
Aider supports 40+ languages through tree-sitter. lean-ctx currently supports 26. For codebases in less common languages, Aider has broader coverage. lean-ctx's language support is actively expanding.
| Language Category | lean-ctx | Aider |
|---|---|---|
| Major (JS/TS/Python/Rust/Go/Java) | Yes | Yes |
| Common (C/C++/C#/Ruby/PHP/Swift) | Yes | Yes |
| Emerging (Zig, Elixir, Dart) | Partial | Yes |
| Niche (COBOL, Fortran, Verilog) | No | Partial |
When to Use Which
Choose Aider if you...
- Want a complete AI coding assistant (not just context tools)
- Prefer a CLI-based workflow with inline editing
- Need repo-map for 40+ languages
- Value SWE-bench proven performance
- Don't need the repo-map in other AI tools
Choose lean-ctx if you...
- Use Cursor, Claude Code, Codex, or other MCP-compatible agents
- Want PageRank repo-map in your existing workflow (without switching to Aider)
- Need compression, memory, and code intelligence alongside repo-map
- Run multi-agent workflows where context needs to be shared
- Want real-time observability of context window usage
Use Both
lean-ctx and Aider can coexist. lean-ctx supports Aider as an MCP client (lean-ctx init --agent aider). You can use Aider with lean-ctx providing additional context tools — including using lean-ctx's repo-map as a complement to or replacement for Aider's built-in one.
Summary
Aider deserves credit for pioneering PageRank-based repo maps in AI coding — it's a proven concept that significantly improves AI agent performance. lean-ctx brings this same capability to the broader MCP ecosystem, making it available to 28 agents instead of just one.
If you're an Aider user, lean-ctx can enhance your workflow with additional compression and memory tools. If you use other AI coding tools, lean-ctx gives you access to PageRank repo-maps that were previously Aider-exclusive.
Aider is an excellent AI coding tool. We recommend trying both and choosing what fits your workflow.
Get started with lean-ctx | Aider on GitHub | Aider repo-map docs