MinishLab/semble
Coding · tested 2026-10-01 · by the Hlido desk, not the vendor
In short: Code search purpose-built for agents, not humans — the ~98%-fewer-tokens-than-grep claim is the whole thesis, and it is the right thesis.
4 PASS · 1 FAIL of 5 public-surface claims
Quick answer
MinishLab/semble scores 85/100 (STEADY) on Hlido’s independent, hands-on test (reviewed 2026-10-01). STEADY (85) because the tool is tightly designed for a real, expensive problem (agent token spend on code navigation), ships through the right channels (PyPI + MCP server), and documents sensible mechanics (cached, auto- Pricing was not findable on the public surface when tested.
Semble is one of the clearest examples of a tool designed for the agent consumer rather than the human one. Instead of an agent burning context on grep-and-read across a repo, it asks a natural-language question and gets back only the relevant snippets, from a cached embedding index that rebuilds on first run and invalidates automatically when files change. The headline — roughly 98% fewer tokens than grep+read — is a token-economics argument, and token economics is exactly where coding-agent cost and latency actually live, so this is a well-aimed product. It ships as a PyPI package with an MCP server, which is the correct distribution for something meant to be wired into Claude Code, Cursor and similar. This is a public-surface and repository read: the token-reduction and accuracy figures are the vendor's and were not independently benchmarked here, so a buyer should confirm them against their own codebase — but the design is sound and the agent-fit is unusually high.
Why STEADY
STEADY (85) because the tool is tightly designed for a real, expensive problem (agent token spend on code navigation), ships through the right channels (PyPI + MCP server), and documents sensible mechanics (cached, auto-invalidating index). Not VITAL on this pass because the central ~98% token-reduction and accuracy claims are vendor figures unverified by hands-on benchmarking in this review.
Public-surface checklist
- PASS Homepage loads (required)
- PASS Primary value prop (required)
- PASS Cta present (required)
- FAIL Pricing or access
- PASS Evidence or demo
What we saw
1 screenshot captured by the Hlido engine during the reviewed run (run-68ee1df60c18a6c2-github-com). Our own captures — not vendor marketing material.
What it does well
- Built for the agent consumer: natural-language code queries return only relevant snippets
- Ships an MCP server — wires directly into MCP-aware coding agents
- Index is cached on first run and auto-invalidated on file change
- Token-economics framing targets the real cost centre of coding agents
- Distributed on PyPI with an interactive installer for fast setup
What it fails at
- Headline ~98% token-reduction and accuracy claims are vendor figures, not independently benchmarked here
- First-run index build adds latency and compute before the first query
- Value depends on embedding/retrieval quality, which varies by codebase
- Narrow by design — it is a search primitive, not a full coding agent
Best for
- Coding agents and MCP-aware IDEs that need cheap, relevant code retrieval
- Large repositories where grep+read blows the context budget
- Developers wiring semantic code search into agent workflows
Not recommended for
- Users wanting a full coding agent rather than a retrieval component
- Tiny repositories where grep is already cheap enough
- Buyers who cannot pilot the token/accuracy claims on their own code
Pricing & access
- Pricing findable on the public surfaceFAIL
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 2026-06-15.
Compared to
-
Sourcegraph Cody
retrieval-primitive
Cody is a full code-AI assistant with search among many features; Semble is a focused retrieval primitive for agents. Choose Semble when you want a token-efficient search component to compose, Cody when you want an assistant.
-
Aider
context-retrieval
Aider is a coding agent that edits code; Semble feeds a coding agent the right context cheaply. They are complements more than competitors — Semble upstream of a tool like Aider.
Agent relevance
CLI MCP SDK Behavioral-testable
Semble exposes an MCP server and a CLI, so coding agents consume it directly as a retrieval tool. It is purpose-built to be driven by an agent rather than a human, which is the highest-fit integration pattern in this corpus.
Agent-friendly score: 9/10
Score over time
The longitudinal record — every point is the score as published on that date. Raw series.
Evidence
- Fast and accurate code search for agents, ~98% fewer tokens than grep+read — source (2026-10-01) verified
- Natural-language queries return only relevant code snippets — source (2026-10-01) verified
- Indexes are cached on first run and auto-invalidated when files change — source (2026-10-01) verified
- Distributed on PyPI with an MCP server — source (2026-10-01) verified
scorecard.json · transparency passport · registry · methodology
Embed this trust badge
Markdown
[](https://hlido.eu/check/?agent=minishlab-semble)HTML
<a href="https://hlido.eu/check/?agent=minishlab-semble"><img src="https://hlido.eu/badge/minishlab-semble.svg" alt="Hlido trust score"></a>