Tentra

@rdanieli

Visit Server

14 days ago

Tentra gives your AI coding agent memory. The code-graph indexer walks your repository with Tree-sitter locally, extracts symbols and call edges, and stores them in a persistent queryable graph. On the tentra monorepo we measured 99.4% token reduction (156.8× ratio) across 8 "where is X implemented?" queries — 114,644 tokens via file re-read vs 731 tokens via graph query. Uses the "agent-as-LLM" pattern — the Claude/Cursor/Codex agent you already have does the semantic extraction, so there is zero LLM cost on Tentra's side and no API key setup for end users. Also includes an architecture workspace with 9 MCP tools for diagrams and 14-framework code export.

Overview Tools Comments

Overview

Tentra — Memory for AI Coding Agents

Tentra is the persistent memory layer for AI coding agents. Two pillars:

Code graph — Index your repo once. Agents query a structured graph of files, symbols, imports, and call edges instead of re-grepping source every session.
Architecture workspace — Describe a system in natural language, get an interactive diagram and production-ready code in 14 frameworks.

Why it saves tokens

Dogfood benchmark on the Tentra monorepo:

99.4% token reduction (156.8× ratio) across 8 "where is X implemented?" queries
114,644 tokens via file re-read → 731 tokens via graph query
Zero LLM cost on Tentra's infra (agent-as-LLM pattern: your Claude/Cursor/Codex agent does the semantic extraction)
Zero API key setup on your side

Setup

Option 1: SSE (zero install)

Cursor / Claude Code / Codex / Windsurf — add to your MCP config:

```json { "mcpServers": { "tentra": { "type": "sse", "url": "https://trytentra.com/api/mcp?key=YOUR_API_KEY" } } } ```

Get your API key at https://trytentra.com/settings after a one-click GitHub sign-in.

Option 2: Local (stdio)

```bash npx -y tentra-mcp ```

Authenticates via GitHub on first use. Needed for local codebase scanning.

32 MCP Tools

Architecture (9): create_architecture, update_architecture, get_architecture, list_architectures, analyze_codebase, lint_architecture, sync_architecture, export_architecture, create_flow
Code Graph — Write (4): index_code, index_code_continue, record_semantic_node, get_index_job
Code Graph — Read (10): query_symbols, get_symbol_neighbors, get_service_code_graph, explain_code_path, find_similar_code, record_embedding, list_god_nodes, get_quality_hotspots, list_snapshots, diff_snapshots
Enrichment (9): set_service_mapping, set_domain_membership, record_contract, bind_contract, record_decision, link_decision, get_ownership, get_decisions_for, get_contracts

What agents use it for

"Where is auth handled?" → query_symbols → file paths + line ranges in one call
"What calls this function?" → get_symbol_neighbors → BFS through the call graph
"Why was this service split?" → get_decisions_for → linked ADRs
"Who owns this file?" → get_ownership → team / person
"Find code similar to this snippet" → find_similar_code → pgvector cosine ANN

Try in Playground

Server Config

{
  "mcpServers": {
    "tentra": {
      "type": "sse",
      "url": "https://trytentra.com/api/mcp?key=YOUR_API_KEY"
    }
  }
}

Build with ShipAny.