| name | mcp-code-intelligence |
| description | Compare MCP code-intelligence layers — CLI wrappers, ast-grep, symbol/graph tools, Zoekt, embeddings. Use when choosing search/dispatch vs heavyweight indexes. Backend domain MCP is image-scoring-mcp (is-be-mcp). |
MCP code intelligence
Compare approaches for giving coding agents repository awareness.
Purpose
Choose the lightest effective layer: CLI tools first, compact domain MCP second, graph/embedding indexes last.
When to use
- Evaluating whether to add an MCP server vs using
rg/fd
- Debugging agent search quality or memory use
- Complementing (not replacing) backend
image-scoring-mcp
Required tools
Depends on tier — baseline CLI from agent-cli-hub.
Recommended tiers
Minimal MCP setup
rg + fd + read_file + git diff + patch_file
Lowest memory, transparent, deterministic. Best default for most tasks.
Better setup
fff-be MCP (grep, find_files, multi_grep) + rg + fd + ast-grep + is-be-mcp search/dispatch + WSL pytest/doctor
Adds indexed file search for long agent sessions and compact pipeline triage via is-be-mcp. Keep rg/fd for one-off shell probes.
Advanced setup
Graphify (graphify query/path/explain or optional graphify-be MCP) + optional Serena / Zoekt / embeddings
Higher setup cost. Prefer Graphify for local AST knowledge graphs (no vector store) before heavier symbol/embedding indexes. Use when cross-module “why / how connected” questions outgrow rg/fff.
Warning: Embedding-first indexing is often heavier and less exact than text/structural search — keep it secondary. Do not use Graphify for pipeline/DB triage — that stays on is-be-mcp.
Comparison matrix
| Layer | Examples | Strengths | Cost |
|---|
| Indexed file search | fff MCP (grep, find_files, multi_grep) — project fff-be | Warm index, frecency, git-aware | Project MCP install |
| CLI wrappers | rg, fd, bat, git diff | Fast, bounded, no index | Agent orchestrates |
| Domain compact MCP | is-be-mcp search/dispatch | Pipeline, DB, diagnostics registry | Build mcp-server/ |
| Structural | ast-grep, semgrep | Syntax shapes | Medium |
| Knowledge graph | Graphify (graphifyy CLI; optional graphify-be MCP) | AST edges, path/query/explain; local, no vectors | uv tool install; build graphify-out/ |
| Graph / embeddings | Serena, Zoekt, claude-context | Symbols, fuzzy discovery | Heavy |
Backend domain MCP
For Vexlum Scoring pipeline, DB, and diagnostics — use compact search + dispatch on is-be-mcp (stdio) and optional is-be-live (SSE when WebUI runs).
See image-scoring-mcp and AGENTS.md.
Gallery triage: sibling workspace is-ui-mcp.
Agent-safe patterns
- Start with Minimal tier; escalate only when text search fails repeatedly.
- Side-effecting
dispatch requires confirmed=True per MCP contract.
- Bound MCP tool output; prefer dispatch actions with
limit parameters.
Commands requiring confirmation
- Installing/running new MCP servers that execute shell or network code
- Side-effecting dispatch without user approval
- Embedding index builds over entire monorepo without scope
See commands-requiring-confirmation.md.
Troubleshooting
- MCP load failures:
cd mcp-server && npm install && npm run build
- webui_unavailable: WebUI not running — use stdio
is-be-mcp local actions or start WebUI for SSE.
Verification checklist
test -f mcp-server/dist/compactIndex.js && echo ok
rg --version && fd --version
Tools not verified in this pass: Graphify MCP (graphifyy[mcp]), Serena, codebase-memory-mcp, Zoekt server, claude-context — treat as optional third-party. See AGENTS.md § Graphify.