Give your AI agent
architectural awareness
Softagram's MCP server lets Claude Code, Cursor, and other AI agents query your codebase architecture instantly — dependencies, impact analysis, and security audits in milliseconds instead of hundreds of grep cycles.
The problem: AI agents fly blind
Without architectural awareness
- AI greps for callers — misses indirect dependencies
- Reads files one by one — burns tokens and time
- Can't see dependency cycles or layering violations
- Changes compile but silently break architecture
- No way to assess blast radius before editing
With Softagram MCP
- One call returns all callers across the entire codebase
- Pre-computed model — millisecond queries, minimal tokens
- Cycles, hubs, and boundary violations detected automatically
- Impact analysis before every change — measure twice, cut once
- Security audit across all repos in a single call
Dramatically reduce token usage
Every grep/read cycle burns tokens and context window. The MCP server replaces hundreds of file operations with structured queries on a pre-computed model.
| Task | Without MCP | With MCP | Savings |
|---|---|---|---|
| Find all uses of a class | 8 greps + 5 file reads | 1 dependency query | ~80% |
| Understand module structure | 20+ file reads | 1 structure query | ~95% |
| Assess change impact | 15 greps + 10 reads | 1 impact analysis call | ~90% |
| Security audit (org-wide) | Manual, per-repo | 1 security audit call | ~99% |
Less token usage means more context window for reasoning — leading to better results in both vibe coding and AI-assisted software development.
8 tools your AI agent gets
sgraph_analyze_change_impact
Before changing code, see every caller and affected component. Warns about dependency cycles and hub elements.
sgraph_get_element_dependencies
Query incoming or outgoing dependencies at any abstraction level — function, file, directory, or repository.
sgraph_search_elements
Find classes, functions, and files by pattern. Replaces grep with structured, type-filtered results.
sgraph_get_element_structure
Explore code hierarchy without reading source files. Understand module layout in one call.
sgraph_security_audit
6-dimension security scan: secrets, CVEs, outdated frameworks, code risk, ownership, and bus factor.
sgraph_audit
Architectural health check: find dependency cycles, hub modules, and layering violations.
sgraph_load_model
Load a pre-computed architecture model. Supports 30+ languages, handles millions of elements.
sgraph_resolve_local_path
Map graph paths back to filesystem paths when you need to read the actual source code.
How it works
Analyze
Softagram analyzes your codebase (30+ languages) and produces an architecture model file.
Load
The MCP server loads the model into memory. Millisecond queries on millions of elements.
Connect
Add the MCP server to Claude Code, Cursor, or any MCP-compatible AI tool.
Build better
Your AI agent now sees the full architecture. Fewer broken builds, cleaner code, lower costs.
Open source
The MCP server and the sgraph data library are open source. The analyzer that produces the models is available as a cloud service or self-hosted deployment.
MCP server for AI agents
Architecture graph library (PyPI)
Use cases
Vibe Coding
Build fast without breaking things. Your AI agent checks blast radius before every change, catching architectural violations that would otherwise ship silently.
AI-Assisted Development
Planning phase gets dramatically cheaper. The agent queries structure and dependencies instead of reading hundreds of files, leaving more context window for actual reasoning.
Security and Compliance
One call scans your entire organization: leaked secrets, known CVEs, EOL frameworks, code quality risk, ownership gaps, and bus factor — across all repositories.
Stop your AI agent from flying blind
Get the architecture model for your codebase. 14-day free trial, 30+ languages supported.