ohwise-mcp

MCP server connecting AI coding agents to graph-native code context and multi-agent pipelines.

ohwise-mcp implements the Model Context Protocol so that any MCP-compatible AI agent (Claude Code, Claude Desktop, and others) can:

• Build and query knowledge graphs from code repositories via codebase2graph
• Build and query knowledge graphs from documents via docs2graph
• Rank the most relevant nodes for any query using Personalized PageRank
• Trigger and poll OhWise Studio pipelines for multi-agent task execution

Pure Python. No LLM dependency for graph tools. Bring your own model.

---

Quick start

pip install ohwise-mcp[all]

Add to Claude Code

claude mcp add ohwise -- ohwise-mcp

Or manually in your Claude Code config (~/.claude.json or .mcp.json):

{
  "mcpServers": {
    "ohwise": {
      "command": "ohwise-mcp",
      "env": {
        "OHWISE_URL": "https://your-ohwise-instance.com",
        "OHWISE_TOKEN": "your-token-here"
      }
    }
  }
}

OHWISE_URL and OHWISE_TOKEN are only required for Studio pipeline tools. Graph tools work offline without them.

---

Tools

Code graph tools

Tool	Description
build_code_graph(repo_path, graph_type)	Extract a knowledge graph from a code repository
rank_code_nodes(query, graph_id, k)	Rank nodes by relevance to a query — get focused code context
search_code_graph(keyword, graph_id, kind_filter)	Find nodes by keyword or kind (function, class, file, …)

Graph types: all, call, entity, schema, workflow, infra, security, web, android, decision, folder

Document graph tools

Tool	Description
build_doc_graph(paths, graph_type)	Extract a knowledge graph from documents (PDF, DOCX, MD, HTML, …)
rank_doc_nodes(query, graph_id, k)	Rank document nodes by relevance — get focused document context

OhWise Studio tools (requires OHWISE_URL + OHWISE_TOKEN)

Tool	Description
start_pipeline(user_input, agent_ids)	Trigger an OhWise coordinator pipeline
get_pipeline_result(thread_id, poll_seconds)	Poll a pipeline for results

---

Example usage in Claude Code

Once configured, Claude Code can use graph context automatically:

> What does the authentication flow look like in this repo?

Claude Code calls build_code_graph → rank_code_nodes("authentication flow") → receives ranked nodes with call relationships and content snippets → answers with precise, relationship-aware context.

> Delegate this refactoring task to the OhWise pipeline and get back the plan

Claude Code calls start_pipeline → OhWise native agents run in parallel → Claude Code receives the synthesized result.

---

Installation variants

# Graph tools only (no OhWise backend needed)
pip install ohwise-mcp[code]    # code graphs
pip install ohwise-mcp[docs]    # document graphs
pip install ohwise-mcp[all]     # both

# Core only (Studio pipeline tools work without graph extras)
pip install ohwise-mcp

---

Python API

The tools are also importable directly:

from ohwise_mcp.server import mcp

# Run as MCP server
mcp.run()

---

Design principles

• No LLM dependency — graph construction is pure static analysis
• Offline-first — code and document graph tools work without any server
• Composable — works standalone or alongside OhWise Studio
• Model-agnostic — any MCP-compatible agent can use these tools

---

Related projects

Package	What it does
codebase2graph	Code repository → knowledge graph
docs2graph	Documents → knowledge graph
graph2sql	Schema graph → SQL context

---

Contributing

git clone https://github.com/jw-open/ohwise-mcp
cd ohwise-mcp
pip install -e ".[dev]"
pytest tests/ -v

---

License

Apache-2.0 — see LICENSE