data-ai MCP Server
Repo memory for coding agents. Local-first MCP for Claude Code, Cursor, Windsurf, and Codex CLI: symbol graph, blast radius, diff-aware review, and git-pinned decisions. MIT; no API keys or code upload.
Discovered via github-topic:model-context-protocol and last synced 3mo ago.
1. Install the package
npx sverklo
2. Add to claude_desktop_config.json
{
"mcpServers": {
"sverklo": {
"command": "npx",
"args": [
"sverklo"
]
}
}
}Config file location: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) / %APPDATA%\Claude\claude_desktop_config.json (Windows)
$30/dev/mo
Location
proprietary
✓
Files
> **First 5 minutes:** see [`FIRST_RUN.md`](FIRST_RUN.md) for three scripted prompts that demonstrate the tools sverklo adds that grep can't replace. --- ## Grep vs Sverklo — the same question, side by side Every one of these is a query a real engineer asked a real AI assistant last week. Grep gives you lines. Sverklo gives you a ranked answer.
Parallel multi-channel fan-out (FTS / vector / path / symbol) with per-channel RRF
Query symbols tagged with a design pattern (observer, repository, validator, ...)
Structural pattern matching across the AST, not just text
**Lint your codebase for AI-readiness.** God nodes, hub files, dead code, circular deps, security smells, A-F health grade — all in one call
Save decisions, patterns, invariants — pinned to the current git SHA
Slice a stored response by line range
F1
$0
MCP
Browse the LLM-derived concept index (themes across the codebase)
Semantic search over saved memories with staleness detection
Grep within a stored context window
0.25
yes
Hybrid BM25 + ONNX vector + PageRank, fused with Reciprocal Rank Fusion
Natural-language router — concepts + investigate + refs in one call
Which tests cover which changed symbols; flag untested production changes
Index health check, file counts, last update
Local
Auto-setup
Semantic clusters of related symbols, computed offline
List all memories with health metrics (still valid / stale / orphaned)
Token-budget stats for stored response handles
0.29
yes
Structural codebase map ranked by PageRank importance
Semantic search restricted to the changed surface of a diff
500-token codebase summary for system prompts on agents that can't run MCP
yes
Walk the symbol graph, return ranked transitive callers (the real blast radius)
Find any function, class, or type by name (typo-tolerant)
Find all references to a symbol, with caller context
Verify a quoted code span is still present at the cited SHA — citation gate
Take the first N hits from the previous response
Risk-scored review of `git diff` — touched-symbol importance x coverage x churn
One-call onboarding — combines overview, code, and saved memories
File dependency graph — both directions, importers and imports
What
Peek at a referenced span by its handle without expanding it fully
10 s
Second-pass critique of a review — what did the first read miss
Delete a memory
0.30
$0
no
yes
yes
Wider candidate pool with refinement hints between rounds
Grep inside the previous result block instead of re-running the search
Query symbols tagged with a design pattern (observer, repository, validator, ...)
Semantic search restricted to the changed surface of a diff
Peek at a referenced span by its handle without expanding it fully
Hybrid BM25 + ONNX vector + PageRank, fused with Reciprocal Rank Fusion
Find any function, class, or type by name (typo-tolerant)
**Lint your codebase for AI-readiness.** God nodes, hub files, dead code, circular deps, security smells, A-F health grade — all in one call
List all memories with health metrics (still valid / stale / orphaned)
Index health check, file counts, last update
MR review
Walk the symbol graph, return ranked transitive callers (the real blast radius)
Structural pattern matching across the AST, not just text
Risk-scored review of `git diff` — touched-symbol importance x coverage x churn
Browse the LLM-derived concept index (themes across the codebase)
Second-pass critique of a review — what did the first read miss
Grep inside the previous result block instead of re-running the search
Wider candidate pool with refinement hints between rounds
Verify a quoted code span is still present at the cited SHA — citation gate
Slice a stored response by line range
3.2
Find all references to a symbol, with caller context
One-call onboarding — combines overview, code, and saved memories
Delete a memory
500-token codebase summary for system prompts on agents that can't run MCP
Parallel multi-channel fan-out (FTS / vector / path / symbol) with per-channel RRF
Grep within a stored context window
Natural-language router — concepts + investigate + refs in one call
Save decisions, patterns, invariants — pinned to the current git SHA
Token-budget stats for stored response handles
Structural codebase map ranked by PageRank importance
File dependency graph — both directions, importers and imports
Semantic search over saved memories with staleness detection
Semantic clusters of related symbols, computed offline
Which tests cover which changed symbols; flag untested production changes
Take the first N hits from the previous response