The context layer for AI coding agents

Give AI agents a map
before they explore.

enola. is an open-source MCP server that generates structured architectural snapshots — modules, symbols, dependencies, routes — so Claude Code, Cursor, Copilot and Codex skip exploration and start with the map.

StatusOSS · pre-1.0 ProtocolMCP LicenseApache 2.0
without enola.
~219K tokens · 68s
$ claude "Where is auth applied?"
 list_files src/                  2.1K tokens
 read src/middleware/auth.ts      4.7K tokens
 read src/server/index.ts         6.2K tokens
 grep -r "requireAuth" .          8.4K tokens
 read src/routes/api/users.ts     5.1K tokens
 read src/routes/api/billing.ts   6.8K tokens
 … 64 more tool calls
with enola.
~5K tokens · 5s
$ claude "Where is auth applied?"
 enola.snapshot(scope=auth)     ~5K tokens
 Auth applied at src/middleware/auth.ts
  mounted by src/server/index.ts:42
  wraps 47 routes across 6 modules
  bypassed by 3 public endpoints
 14 scoped lookups — all served from the map
Works with every MCP-compatible agent
One config line. Local or hosted, same install.
Claude Code Cursor GitHub Copilot Codex & more

The problem

Three costs hiding inside
every AI-assisted PR.

Your agents are powerful — but they're blind. Every session starts from zero: re-crawling files, re-grepping symbols, re-learning structure.

01 · Tokens
~219Ktokens

Burned exploring structure.

A typical multi-agent code-understanding task — all exploration, nothing shipped — before a single change is made.

Per task · per repo · per session
02 · Time
~68sper task

Wasted on exploration latency.

Multiply by every PR, every developer, every day — exploration latency compounds silently across the org.

Compounding across the org
03 · Cost
~90%of spend

Burned before any real work.

The overwhelming majority of every token bill goes to re-exploration that produces no change — on every task, in every repo. A shared map recovers it.

Recoverable spend

The solution

One architectural map.
Every agent uses it. Every session.

enola. analyzes your repository once and exposes a compact, structured graph of its architecture through the Model Context Protocol — so any MCP-capable agent gets oriented in milliseconds, not minutes.

What enola. delivers to agents

  • Cross-repo linking & incremental updates
  • Pattern detection (auth, data flow, queues)
  • Go · TypeScript · Python · Kotlin · Swift · Ruby · OpenAPI
  • Delivered through any MCP-compatible agent

The outcome

  • Agents skip exploration — they get the map first
  • Token spend collapses across every workflow
  • Output quality goes up — less hallucinated architecture
  • Every change is attributable and reviewable

The benchmark

Same understanding.
~44× fewer tokens. ~14× faster.

One representative repository. One architectural-understanding task. enola. versus an explore-and-read multi-agent baseline.

Metric enola. MCP Explore agents Delta
Tool calls 15 70 across 3 agents −79%
Tokens used ~5K est. ~219K ~44× less
Time to context ~5s ~68s ~14× faster
Facts extracted 102,058 N/A — narrative only structured

Measured with Claude Haiku · 102,058 extracted facts · 3 repositories. Results vary by model, codebase size, and repository structure.

Change intelligencePaid plan

See what changed
and what it cost you.

A paid add-on that maps every commit against the architectural graph: which agent touched which module, the dependencies it introduced, its blast radius. Reviewable, exportable, auditable. We're onboarding teams by hand — reach out to scope it.

What you'll see

  • New dependencies introduced by AI changes
  • Module movement & API-surface drift
  • Agent vs. human commit volume per branch
  • Token spend by repo, team, and tool

What it changes

  • Catch unsafe AI changes before merge
  • Re-target spend toward high-leverage repos
  • Measure delivery velocity, attributable to AI
Repository activity · last 24h
00:00 08:00 16:00 24:00 main develop feat/auth feat/bill ai/sweeps
15,247
commits / day
38%
by AI agents
Human AI agent