How to Install mcp-repo-graph in OpenAI Agents SDK
Structural graph memory for AI coding assistants navigate any codebase by structure, not guesswork.
What mcp-repo-graph does
repo-graph gives AI coding assistants a structural map of your codebase — entities, relationships, and end-to-end feature flows — so they navigate straight to the files that matter instead of grepping and reading everything first. It scans your repo once and builds a lightweight graph of what exists (modules, classes, functions, routes, services, components), how things connect (imports, calls, handles, cross-stack HTTP), and where each feature begins and ends. The assistant queries that graph through 11 MCP tools (status, flow, trace, impact, activate, find, dense_text, graph_view, …), finds the minimal set of files for the task, and reads only those. Works across 20+ languages and frameworks (Go, Rust, TypeScript/React/Angular/Vue, Python, Java, C#, and more), with cross-stack linking between frontend calls and backend routes. The engine is native Rust shipped as a prebuilt wheel, so scans are fast — a 12,000-node repo maps in a couple of seconds. Controlled before/after on a Go+Angular bug fix, same model and prompt: 2.5× fewer tokens and ~9× faster with repo-graph installed.
Installation steps
- 1Install: pip install openai-agents
- 2Copy the snippet below into your agent code
- 3Replace placeholder env values with your real keys
- 4Pass mcp_servers=[mcp_server] to your Agent()
Python snippet
from agents import Agent
from agents.mcp import MCPServerStdio
mcp_server = MCPServerStdio(
command="uvx",
args=["mcp-repo-graph", "--repo", "/path/to/your/project"],
)
agent = Agent(
name="My Agent",
model="gpt-4o",
mcp_servers=[mcp_server],
)Install in other clients
For maintainers
If you maintain mcp-repo-graph, add this badge to your README to show it's verified on CuratedMCP:
[](https://curatedmcp.com/marketplace/mcp-repo-graph)