What did Hlido score codebase-memory-mcp?

codebase-memory-mcp scored 82/100 (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.

codebase-memory-mcp

Name: codebase-memory-mcp review
Item: codebase-memory-mcp
Rating: 82
Author: Hlido Editor

Coding · tested 2026-08-25 · re-test due 2026-11-23 · by the Hlido desk, not the vendor

In short: A structural-analysis backend for coding agents — indexes a repo into a persistent knowledge graph so the agent answers structural questions with far fewer tokens, and it is honest that it holds no LLM of its own.

5 PASS · 0 FAIL of 5 public-surface claims

Visit codebase-memory-mcp → scorecard.json

Quick answer

codebase-memory-mcp scores 82/100 (STEADY) on Hlido’s independent, hands-on test (reviewed 2026-08-25). STEADY (82) because this is a technically serious, open-source, agent-native tool with an unusually honest scope statement (no embedded LLM, no API key) and — rare for the category — a research preprint backing its effic

Hlido Editor · 2026-08-25

codebase-memory-mcp is an open-source MCP server that turns a repository into a persistent knowledge graph of functions, classes, call chains, HTTP routes and cross-service links, so an AI coding agent queries the graph instead of reading files one at a time. The pitch is efficiency: Tree-sitter parsing across 158 languages, Hybrid LSP type resolution, a native executable that needs no language runtime, and — the headline number — roughly 120x fewer tokens on structural questions. The most important sentence on the whole page is a disclaimer of ambition: 'It is a structural-analysis backend, not a chatbot: there is no embedded LLM and no API key. Your MCP client is the intelligence layer.' That honesty about what it is not is exactly what you want from infrastructure, and it is rarer than it should be. The project also does something most tools in this category do not: it grounds its claims in a research preprint (arXiv:2603.27277) with a stated evaluation across 31 real-world repositories — 83% answer quality, 10x fewer tokens, 2.1x fewer tool calls versus file-by-file exploration. That is a credibility multiplier, with the usual caveat that these are the vendor's own benchmarks and the two token figures quoted on the page (120x in the hero, 10x in the preprint summary) are not the same number and a buyer should read the methodology before quoting either. Net: a technically serious, honestly-scoped tool that a token-conscious coding agent can genuinely benefit from, held just short of the mid-80s only by the gap between the hero's 120x and the preprint's 10x and the absence of independent verification.

Why STEADY

STEADY (82) because this is a technically serious, open-source, agent-native tool with an unusually honest scope statement (no embedded LLM, no API key) and — rare for the category — a research preprint backing its efficiency claims across 31 repositories. It is held below the mid-80s because the two token-reduction figures on the surface (120x hero vs 10x preprint) are inconsistent without reading the methodology, and all benchmarks are vendor-run and not independently reproduced.

Public-surface checklist

PASS Homepage loads (required)
PASS Primary value prop (required) — The fastest, most efficient code intelligence engine for AI coding agents
PASS Cta present (required) — View on GitHub / Download latest release
PASS Pricing or access — Open-source, free; native executable download and GitHub source with no API key required
PASS Evidence or demo — Research preprint (arXiv:2603.27277) with benchmarks across 31 repos; 3D graph visualisation; documented tool surface

What we saw

4 screenshots captured by the Hlido engine during the reviewed run (run-ff0d9e34f3effd18-deusdata-github-io). Our own captures — not vendor marketing material.

codebase-memory-mcp — run screenshot 1 (home.png) — home.png

codebase-memory-mcp — run screenshot 2 (page_what-is-it.png) — page_what-is-it.png

codebase-memory-mcp — run screenshot 3 (page_install.png) — page_install.png

codebase-memory-mcp — run screenshot 4 (page_hybrid-lsp.png) — page_hybrid-lsp.png

What it does well

Indexes a repo into a persistent knowledge graph of functions, classes, call chains, routes and cross-service links
Explicitly honest about scope: 'a structural-analysis backend, not a chatbot ... no embedded LLM and no API key'
Tree-sitter parsing across 158 languages with Hybrid LSP type resolution
Ships as a native executable requiring no language runtime — low install friction
Grounds claims in a research preprint (arXiv:2603.27277) evaluated across 31 real-world repositories
Open-source MCP server that works with Claude Code and any MCP-compatible agent
Optional 3D graph visualisation for humans to inspect the same graph the agent queries

What it fails at

The two headline token figures disagree — 120x fewer tokens in the hero versus 10x in the preprint summary — without a reconciling note on the surface
All efficiency and quality benchmarks are vendor-run; none are independently reproduced
Value depends on the agent's structural questions matching what the graph indexes; free-text semantic recall is a separate feature
Young project; no continuity, maintenance-cadence or funding signal published
No named production adopters beyond the preprint's evaluation repositories

Red flags

Inconsistent token-reduction claims on the same surface (120x in the hero, 10x in the cited preprint) — read the methodology before quoting either figure

Best for

Token-conscious coding agents that answer structural questions (callers, routes, impact) over large repos
Teams indexing very large codebases where file-by-file exploration is too slow or too expensive
MCP users who want a runtime-free native backend with no API key to manage

Not recommended for

Users wanting an all-in-one coding assistant — this is a backend, not the intelligence layer
Buyers who need independently verified benchmarks before adopting
Workflows dominated by free-text semantic search rather than structural queries

Pricing & access

Pricing findable on the public surfacePASS Open-source, free; native executable download and GitHub source with no API key required (tested 2026-08-25)

Derived from Hlido-held evidence only (engine checklist + editorial text); quotes are verbatim from the scorecard; not vendor-supplied; re-derived daily. Verify current prices on the vendor's pricing page. Last verified 2026-08-25.

Compared to

Serena structural-graph-vs-lsp-navigation
Serena is a semantic code-navigation MCP toolkit that leans on LSP for symbol-level operations; codebase-memory-mcp builds a persistent structural knowledge graph across 158 languages with a research-preprint backing. Serena for LSP-driven symbol work, codebase-memory-mcp for graph-scale structural queries with token efficiency as the headline.
Neo4j MCP Integrations purpose-built-code-graph
Neo4j's MCP server exposes a general graph database to agents; codebase-memory-mcp is a purpose-built code-structure graph that indexes repos automatically. Neo4j's for arbitrary graph data you model yourself, codebase-memory-mcp for out-of-the-box codebase intelligence.

Agent relevance

CLI MCP Behavioral-testable

Agentic-Commerce Readiness 64/100 · INTEGRABLE

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

A native MCP server. A coding agent (Claude Code or any MCP client) connects to codebase-memory-mcp and queries the pre-built knowledge graph — callers, routes, impact, cross-service links — instead of reading files. No API key or embedded model; the agent supplies the reasoning and the server supplies fast structural facts. Runtime-free native binary keeps setup minimal.

Agent-friendly score: 9/10

Evidence

Open-source MCP server indexing a codebase into a persistent knowledge graph queried by the agent — source (2026-08-25) verified
Explicitly holds no embedded LLM and requires no API key; the MCP client is the intelligence layer — source (2026-08-25) verified
Tree-sitter parsing across 158 languages, native executable, Hybrid LSP type resolution — source (2026-08-25) verified
Efficiency claims backed by a preprint evaluated on 31 repositories, but hero (120x) and preprint (10x) token figures differ — source (2026-08-25)

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-1+editorial-narrative-v2 · Methodology version 2026.05 · Next review due 2026-11-23

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/deusdata-codebase-memory-mcp.svg)](https://hlido.eu/check/?agent=deusdata-codebase-memory-mcp)

HTML

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