MCP Integrations

Connect SourcePrep to your favorite AI editors using the Model Context Protocol.

Agentic IDEs

Cursor, Windsurf, Antigravity, VS Code (via Copilot), and Zed — plus MCP-aware VS Code extensions like Cline.

CLI Agents

Claude Code (primary), OpenAI Codex, and Gemini CLI — connect any MCP-aware command-line agent directly to your local SourcePrep index.

Paperclip Plugin

New

Give every Paperclip agent structural codebase intelligence. 5 tools, dashboard widgets, and event-driven context — installed as a native Paperclip plugin.

What is MCP?

The Model Context Protocol (MCP) is an open standard that enables AI models to interact with external data and tools.

SourcePrep runs a local MCP server that exposes your indexed codebase as a set of tools. It supports both Stdio (recommended for local editors) and SSE (for remote/containerized setups). When you connect a CLI agent like Claude Code or Codex, or an IDE like Cursor or Windsurf, their internal AI agents gain the ability to:

Get Oriented: Call prep with no arguments for ambient context — module structures, hub files, focus areas, and immune-system alerts.
Search Semantically: Use prep_search to find code by meaning. The query intent is auto-classified (LOCATE, EXPLAIN, RATIONALE, TRACE, EXAMPLE, COMPARE, DISCOVER) so the right index gets queried.
Code Graph: Use prep_impact to see exact blast radius before editing — what depends on a file or symbol, with stdlib/external noise filtered out.
Audit Codebase: prep_audit is dual-mode — call with no args for structural findings (coupling, cycles, concept violations), or pass findings to enrich ruff/eslint/semgrep output with structural context. Accepts SARIF in/out. Learn more →
Persistent Memory: Use prep_observe to save cross-session notes that get automatically flagged as [STALE] when the underlying files change.
Business Rationale: Use prep_concepts to record and query the "why" behind code — design decisions, constraints, and architecture rules. Constraint concepts auto-generate antibodies that surface in ambient context when violated.

Tools Reference

Tool	Purpose
prep	Ambient context — module map, hub files, focus areas, immune-system alerts
prep_search	Semantic search with auto-classified intent (locate / explain / rationale / trace / example / compare / discover)
prep_impact	Blast radius analysis — what depends on a file or symbol, with noise filtered
prep_audit	Structural findings + lint enrichment (action = scan, antibodies, refactor, verify, report, advise)
prep_observe	Save / retrieve cross-session notes with stale flags
prep_concepts	Record / query design decisions and constraints; constraint concepts auto-derive antibodies

claude — my-project

❯

What a prep call looks like in your editor — ambient context, no arguments.

Live Dashboard Preview

These are live, interactive previews of the SourcePrep dashboard panels your agents interact with.

Concept / Question

Results (K)

Min Score

Semantic Search Panel — find code by meaning, not keywords.

~/code/sourceprep

Fresh

10Code

12Docs

18Graph

22Total

Code 25%Docs 30%Graph 45%

Knowledge Status — real-time view of your codebase knowledge.

Engines

HR Agent

Generate and audit agent role definitions

5 pushed · 2 hours ago

Researcher

Mine audit findings, formulate plans

3 pushed · 45 min ago

Custodian

Detect dead code, plan cleanup

1 day ago

Paperclip Integration

Checking...

Install the SourcePrep skill so Paperclip agents can use structural codebase intelligence tools (prep, prep_search, prep_impact).

Push Settings

Auto-push findings to Paperclip

Threshold:

Project:

Agent Operations Panel — the cross-cutting view of which MCP-aware agents are using your index right now.