Beta
20+ MCP tools · 19 languages · Knowledge graphs
Your AI wastes 90% of its tokens reading files
Every file read, grep, and directory listing burns tokens you're paying for. Arc's knowledge graph gives AI coding tools the same understanding in 10x fewer tokens — and a 10x smaller API bill. Hosted in the cloud or as a Mac app.
Every question about your codebase costs thousands of tokens in file reads, greps, and directory listings. Most of that context is wasted — and you're billed for all of it.
Priced at
Without Arc
25,000+
tokens to understand 8 files
$0.075per question
Full file reads for structure
Grep to find definitions
Multiple round trips
Context window fills fast
Repeated exploration per session
→
With Arc
2,500
tokens for the same understanding
$0.008per question
Knowledge graph queries
Semantic search across symbols
Single call per question
More room for reasoning
Persistent graph across sessions
That's $0.068 saved every time your AI asks a question — $67.50 per 1,000 questions at Claude Sonnet 4 input rates.
Tools
20+ tools your AI actually needs
Not just file summaries. Full knowledge graph queries — callers, callees, impact radius, execution flows, communities, and more.
Inventory
list_repos
Connected repositories, indexing status, and desktop rollups
Hive
repo_activity
Paid live repo presence and dirty-file risk counters
Search
semantic_search_nodes
Find functions, classes, and types by name or description
Search
query_graph
Callers, callees, imports, inheritors, tests for any symbol
Search
list_graph_stats
Node counts, edge counts, languages, last update time
Impact
get_impact_radius
Blast radius — what files and functions are affected by a change
Impact
detect_changes
Risk-scored change analysis with test coverage gaps
Impact
get_review_context
Token-optimized context for code review with guidance
Impact
get_affected_flows
Which execution flows are impacted by file changes
Architecture
list_communities
Auto-detected code clusters via graph analysis
Architecture
get_architecture_overview
High-level map of modules and their relationships
Architecture
list_flows
Execution flows sorted by criticality score
Quality
find_large_functions
Spot oversized code that needs refactoring
Quality
generate_wiki
Auto-generated documentation from code structure
+ 6 more tools for flows, communities, embeddings, context, and documentation
How It Works
Three minutes to smarter AI
01
Sign up
Sign in with GitHub. Get your API key instantly. No credit card.
02
Add repos
Browse public GitHub repos, or add private clone URLs after configuring a deploy key.
03
Connect
One config line in Claude Code, Cursor, or OpenCode. Your AI gets 20+ tools.
# 1. Install Arc.app from the .dmg# 2. Open it and pick a workspace folder# 3. Click "Install MCP" — Arc writes the config for the tools you have# 4. Your AI tool restarts with Arc MCP tools served locally# Optional: link to agentcodelab.com to share an org's graph with your team# Settings → Cloud Connection → Sign in
Also works with Windsurf, Cline, Continue, Zed, and any MCP-compatible tool.
Languages
19 languages. Your stack is covered.
TypeScript
JavaScript
Python
Go
Rust
Java
C#
C / C++
Kotlin
Swift
Ruby
PHP
Scala
Dart
Solidity
R
Perl
Lua
Vue SFC
Pricing
Start free. Scale when ready.
No credit card required. Upgrade or cancel anytime.