TempoGraph

Your AI agent finds the right files. Every time.

One MCP tool call gives your coding agent the exact files it needs before making changes. No guessing. No irrelevant context.

$pip install tempograph
GitHub stars PyPI version Glama score License
TempoGraph demo — AI agent finding the right files

The problem

  • Agents guess which files to look at
  • They search by filename and grep for keywords
  • They miss critical dependencies
  • They break things downstream
  • They waste tokens on irrelevant code

The fix

  • TempoGraph builds a real dependency graph
  • Agent calls prepare_context
  • Gets the exact files that matter
  • Triple search: structural + semantic + keyword
  • Adaptive gating: knows when to skip context
24
MCP tools
170+
Languages supported
21ms
Warm query time
4,800+
Tests

Benchmarked on real PRs

Tested on django, flask, httpx, fastapi, requests, and pydantic. Task: predict which files need to change.

Model Without TempoGraph With TempoGraph Improvement
GPT-4o 21.7% F1 27.5% F1 +27%
GPT-4o-mini 19.2% F1 24.5% F1 +28%
qwen2.5-coder:32b canonical benchmark (p=0.049) +18.6%

How it works

your repo ──> tree-sitter parse ──> symbols + edges ──> SQLite graph

    Agent calls prepare_context(task="fix auth bug")

    ──> keyword extraction ──> symbol search ──> graph traversal

    <── KEY FILES + callers + callees + risk signals

When your agent needs more

"What breaks?"

Before changing a file, see every importer and caller that could break.

blast_radius

"Show me everything"

Focus on a symbol and see callers, callees, and related files in context.

focus

"What's risky?"

Find files where complexity, coupling, and size overlap. Change carefully.

hotspots

"What's dead?"

Find exported symbols with zero incoming references. Safe to delete.

dead_code

"What changed?"

Impact analysis of your current diff — breaking-change risk signals.

diff_context

"Find anything"

Hybrid search: structural graph + semantic vectors + FTS5 keywords, fused via RRF.

search_semantic

Set up in 30 seconds

Add to your MCP config (Claude Code, Cursor, Windsurf, or any MCP client):

{
  "mcpServers": {
    "tempograph": {
      "command": "tempograph-server",
      "args": []
    }
  }
}

Give your agent better context

Install TempoGraph and stop guessing which files matter.

GitHub PyPI Glama