What did Hlido score Serena?

Serena scored 84/100 (STEADY) in Hlido's independent, hands-on review.

Does any vendor pay Hlido for placement?

No. Hlido takes no money from the agents it rates — scoring weights stay private and the evidence behind every verdict is public.

Serena

Name: Serena review
Item: Serena
Rating: 84
Author: Hlido Editor

Coding · tested 2026-08-19 · re-test due 2026-11-19 · by the Hlido desk, not the vendor

In short: An agent-first, IDE-grade semantic toolkit that plugs into any MCP client — one of the cleaner examples of tooling designed for the model, not the human.

Visit Serena → scorecard.json

Quick answer

Serena scores 84/100 (STEADY) on Hlido’s independent, hands-on test (reviewed 2026-08-19). STEADY (84) for a mature, MCP-native coding toolkit with a genuinely agent-first design, broad documented client support, and an unusually honest evaluation posture, held at low-medium confidence because the review is su

Hlido Editor · 2026-08-19

Serena provides semantic code retrieval, editing, refactoring and debugging tools that operate at the symbol level rather than on line numbers or raw text, and exposes them to any LLM client over MCP — Claude Code, Codex, Gemini-CLI, Cursor, VS Code, JetBrains and the desktop clients are all listed as supported. The pitch is coherent and, unusually, self-aware: the project states outright that its "end users" are the AI agents that actually call the tools, and it publishes an evaluation where agents run ~20 routine coding tasks with and without Serena and report back. Those testimonials (Opus in Claude Code, GPT in Codex CLI, Copilot CLI on a monorepo) are vendor-run and self-selected, so Hlido treats them as illustrative rather than independent proof. What is verifiable from the surface is the design posture — high-level, symbol-aware abstractions that collapse multi-step text surgery into atomic operations, and honest framing that an LLM still does the actual orchestration. Hlido assessed the documentation site only, not a running session, so the efficiency and reliability claims are unconfirmed hands-on. But the architecture is exactly what the agent-to-agent thesis rewards: structured, machine-first, and standards-based via MCP.

Why STEADY

STEADY (84) for a mature, MCP-native coding toolkit with a genuinely agent-first design, broad documented client support, and an unusually honest evaluation posture, held at low-medium confidence because the review is surface-only (docs site, not a running session) and the headline efficiency/reliability gains rest on vendor-run agent testimonials Hlido did not reproduce. Not VITAL because none of the performance claims were independently verified.

What we saw

4 screenshots captured by the Hlido engine during the reviewed run (run-69fc58ba7b3ed1fd-oraios-github-io). Our own captures — not vendor marketing material.

Serena — run screenshot 1 (home.png) — home.png

Serena — run screenshot 2 (page_01-about_000_intro_html_main-content.png) — page_01-about_000_intro_html_main-content.png

Serena — run screenshot 3 (page_index_html.png) — page_index_html.png

Serena — run screenshot 4 (page_01-about_000_intro_html_.png) — page_01-about_000_intro_html_.png

What it does well

Genuinely agent-first tool design — symbol-level abstractions (renames, references, refactors) built for the model to call, not line-number text surgery
Standards-based integration via MCP; documented support across terminal clients (Claude Code, Codex, Gemini-CLI), IDEs (Cursor, VS Code, JetBrains) and desktop/web clients
Unusually honest framing — it names AI agents as the real end users and publishes an evaluation methodology instead of bare marketing
Clear explanation that an LLM is still required to orchestrate; no overreach about autonomous capability

What it fails at

The headline efficiency and reliability gains rest on vendor-run, self-selected agent testimonials — not independent benchmarks
Surface-only review — Hlido read the documentation site, not a live session, so tool behaviour was not hands-on verified
No hosted option; it is a self-run server you wire into your own client
Value is bounded by the host LLM and MCP client — Serena adds tools but does no work on its own

Best for

Developers running an MCP-capable coding agent (Claude Code, Codex, Cursor) who want IDE-grade semantic navigation and refactoring
Teams working in large or complex codebases where cross-file refactors and reference lookups are error-prone for text-based agents
Anyone wanting a client-agnostic, open toolkit rather than a single-vendor coding assistant

Not recommended for

Users wanting a turnkey hosted coding assistant with no setup
Non-technical users — this is developer tooling wired into a coding agent
Anyone needing independently benchmarked performance guarantees before adopting

Related agents

Agent relevance

CLI MCP Behavioral-testable

Agentic-Commerce Readiness 57/100 · INTEGRABLE

Independent readiness for agent delegation & transaction. How it’s scored · check live

Serena runs as an MCP server that any MCP-capable client launches (via a launch command) or connects to over HTTP. Coding agents then call its symbol-level retrieval/edit/refactor tools directly in natural-language workflows.

Agent-friendly score: 9/10

Evidence

Genuinely agent-first tool design — symbol-level abstractions (renames, references, refactors) built for the model to call, not line-number text surgery — source (2026-08-19) verified
Standards-based integration via MCP; documented support across terminal clients (Claude Code, Codex, Gemini-CLI), IDEs (Cursor, VS Code, JetBrains) and desktop/web clients — source (2026-08-19) verified
Unusually honest framing — it names AI agents as the real end users and publishes an evaluation methodology instead of bare marketing — source (2026-08-19) verified
Clear explanation that an LLM is still required to orchestrate; no overreach about autonomous capability — source (2026-08-19) verified
Hands-on runtime behaviour (executing the tool / a live task) — source (2026-08-19)

scorecard.json · transparency passport · registry · methodology

More: compare agents · best of · developer tools · incident registry

Verdict by Hlido Editor, our automated editorial system · Method: public-surface-tier-2+editorial-narrative-v2 · Methodology version 2026.05 · Next review due 2026-11-19

How this page was produced. The scores, claim verdicts and evidence come from automated hands-on testing of the product’s public surface. The written analysis is drafted by an AI system, and pages publish without a person reviewing each one. Hlido publishes this record and answers for it — tell us if anything here is wrong and we will correct it.

Embed this trust badge

Live, always-current independent score — free to embed on your site or README. No vendor pays for placement.

Markdown

[![Hlido trust score](https://hlido.eu/badge/oraios-serena.svg)](https://hlido.eu/check/?agent=oraios-serena)

HTML

<a href="https://hlido.eu/check/?agent=oraios-serena"><img src="https://hlido.eu/badge/oraios-serena.svg" alt="Hlido trust score"></a>