Skip to main content
Glama

Intelligraph-mini

Local-first MCP graph intelligence server — same RRF hybrid search, multi-hop traversal, source snippets, and rationale nodes as the full Intelligraph platform, but without Docker, web UI, SSO, or chat. Just tools for your AI agent.

Need the full platform? Intelligraph adds Docker, React web UI, chat completions, SSO/PKCE, closed-network deployment, and tuning controls.

Quick start

pip install intelligraph-mini

# In your project directory:
intelligraph-mini --repo-dir .

# Or with MCP config (.mcp.json):
{
  "mcpServers": {
    "intelligraph-mini": {
      "command": "intelligraph-mini",
      "args": ["--repo-dir", "."]
    }
  }
}

First run builds graphify + CRG indexes (~60s). Subsequent runs load cached (~2s). The bundled all-MiniLM-L6-v2 model (87MB) works fully offline — no API calls, no network.

Related MCP server: corpus-rag

Tools

Tool

Description

search(query)

RRF hybrid search (FTS5 + semantic embeddings). Finds symbols by meaning.

node(name, depth=2)

Multi-hop subgraph + source code snippets + rationale notes.

path(from, to)

Shortest path between two symbols in the call graph.

impact(name)

Blast-radius analysis over CALLS/IMPORTS_FROM edges.

local_files(paths)

Read source files from disk.

How it works

  1. Build (first run): graphify update . + code-review-graph build → graphify-out/graph.json + .code-review-graph/graph.db

  2. Snippets: reads source files, stores ~500 char snippets per node in node_snippets table

  3. Search: RRF (Reciprocal Rank Fusion, k=30) blends FTS5 keyword ranking with embedding cosine similarity. Adaptive 50% cutoff returns only genuinely relevant files.

  4. Traversal: BFS with token budget over cached adjacency (scales to 140k edges)

  5. Rationale: surfaces #NOTE/#WHY nodes from graphify's rationale extraction

Requirements

  • Python 3.10+

  • graphifyy and code-review-graph CLIs on PATH (installed automatically as dependencies)

License

MIT

Available Tools

5 tools
impactA

Complete blast radius of changing a symbol. Exhaustive traversal of ALL edge types. Returns every affected file with symbols to check. Use before refactoring. Files not listed do not depend on the target.

Inputs: name (required).

local_filesA

Read full source files from disk. EXPENSIVE. Prefer built-in Read with line ranges from search/node results instead. Use this only when you need a whole file that search/node didn't cover.

Inputs: paths (required), max_bytes.

nodeA

Get a symbol's connections (callers, callees, imports) with file:line ranges. Use AFTER search. Then use built-in Read with those line ranges to get implementation details. Replaces reading whole files — read only the specific line ranges shown.

Inputs: name (required), depth.

pathC

Trace the shortest path between two symbols in the codebase graph.

Inputs: to (required), from (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool has a distinct role: search locates symbols, node shows connections, path finds shortest routes, impact assesses blast radius, and local_files reads source. No two tools overlap in purpose.

    Naming Consistency3/5

    Names are mostly single-word and lowercase, but mix verbs (search) and nouns (node, path, impact, local_files). The underscore in local_files is inconsistent with the other names, though still readable.

    Tool Count5/5

    Five tools is well-scoped for a mini code-graph intelligence server, covering search, exploration, pathfinding, impact analysis, and file reading without redundancy.

    Completeness5/5

    The set covers the full analysis workflow: locate, explore, trace, assess, and read. Instructions to use built-in Read with line ranges fill any need for source access, leaving no obvious gaps.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      MCP server for local RAG over personal notes, PDFs, and documents, enabling plain-English querying and hybrid search with multi-hop context expansion.
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      A local MCP server enabling hybrid search over documents, memory, and knowledge graphs for retrieval-augmented generation, with tools for SQLite, semantic memory, and entity-relationship queries.
      4
      1
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Zero-trust, air-gapped Enterprise GraphRAG MCP server. Build knowledge graphs from local documents and run multi-hop, citation-grounded queries entirely offline with Ollama.
      17
      30 PyPI
      326
      MIT