DEV.co
MCP Servers · MinishLab

semble

Semble is a Python-based code search library designed for AI agents that uses semantic embeddings to find relevant code snippets 98% more token-efficiently than traditional grep-and-read approaches. It indexes and searches entire codebases on CPU in milliseconds with no external dependencies, and integrates directly with agent platforms via MCP protocol.

Source: GitHub — github.com/MinishLab/semble
5.5k
GitHub stars
235
Forks
Python
Primary language
MIT
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
RepositoryMinishLab/semble
OwnerMinishLab
Primary languagePython
LicenseMIT — OSI-approved
Stars5.5k
Forks235
Open issues8
Latest releasev0.4.2 (2026-07-06)
Last updated2026-07-06
Sourcehttps://github.com/MinishLab/semble

What semble is

Semble provides fast semantic code search via embedding-based retrieval (NDCG@10: 0.854) with ~250ms indexing and ~1.5ms query latency on CPU. It respects .gitignore and .sembleignore patterns, supports local and remote (git) repositories, and exposes APIs via Python library, CLI, and MCP server for integration with Claude Code, Cursor, Codex, and other MCP-compatible agents.

Quickstart

Get the semble source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/MinishLab/semble.gitcd semble# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

AI Agent Code Exploration

Enable coding agents (Claude Code, Cursor, Codex) to instantly retrieve relevant code snippets for context-aware suggestions and refactoring without reading entire files or grepping manually.

Token-Constrained LLM Integration

Reduce LLM token consumption in agentic workflows by 98% through targeted snippet retrieval, lowering inference costs and latency when agents need to understand unfamiliar codebases.

Local Development Search Infrastructure

Build custom tooling, IDE plugins, or internal search services that leverage semantic code understanding without relying on cloud APIs, GPUs, or paid services.

Implementation considerations

  • Install via `uv tool install semble` or as Python dependency; automated integration helpers exist for Claude Code, Cursor, Codex, and OpenCode via `semble install` command.
  • Indexes are cached locally (OS cache folder or custom location via SEMBLE_CACHE_LOCATION) and automatically invalidated on file changes; no manual cache invalidation required.
  • Library supports three content types (code, docs, config) independently or combined; choose content strategy upfront to avoid reindexing.
  • MCP server integration requires MCP client configuration per agent; CLI and Python library can be used independently without MCP setup.
  • CPU-only execution means no GPU or external API overhead, but indexing speed and memory footprint should be validated against your largest target codebases.

When to avoid it — and what to weigh

  • Requirement for Sparse/Exact Matching — If your use case requires precise regex or exact-string matching (e.g., finding all function calls to a specific API), traditional grep or AST-based tools will be more reliable than semantic search.
  • Very Large Monorepos at Scale — While Semble is fast, indexing performance and memory characteristics for multi-gigabyte codebases are not documented; consider benchmarking against your actual repo size before committing.
  • Guaranteed Deterministic Retrieval — Embedding-based search can vary slightly across runs or model updates; if you need reproducible, deterministic results for compliance or testing, prefer rule-based or AST approaches.
  • Offline-First with Frequent Repo Updates — Index invalidation happens automatically on file changes, but in high-churn environments with constant rewrites, the overhead of continuous reindexing has not been characterized.

License & commercial use

MIT License. Permissive open-source license allowing commercial use, modification, and distribution with attribution. No warranty or liability limitations unusual for MIT.

MIT license explicitly permits commercial use. No vendor lock-in, no proprietary terms, no per-seat or usage-based fees. However, ensure your LLM agent platform (Claude, Cursor, etc.) licensing allows integration of third-party tools; Semble itself is free to deploy and modify.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationAdequate
License clarityClear
Deployment complexityLow
DEV.co fitGood
Assessment confidenceHigh
Security considerations

No security audit or CVE data provided. Local-only execution (no external APIs) reduces attack surface compared to cloud-based alternatives. User repositories are indexed locally and cached on disk; ensure SEMBLE_CACHE_LOCATION has appropriate filesystem permissions if sensitive code is indexed. No encryption of cached indexes noted; requires review if regulatory compliance (e.g., HIPAA, SOC 2) is required.

Alternatives to consider

Tree-sitter + Treesitter-based AST search

Deterministic, exact structural matching; better for refactoring tools and linters. Slower than semantic search, but no embedding model needed. Overkill if semantic relevance is acceptable.

Embeddings-based search (LangChain + ChromaDB/Pinecone)

Flexible, cloud-scalable alternative using external vector DBs. Slower local indexing, requires API keys or external infrastructure. Better for multi-codebase or shared team search. Higher cost and latency for real-time agent queries.

GitHub Code Search API or Sourcegraph

Purpose-built for large public/enterprise codebases with strong filtering. Cloud-hosted, requires internet and authentication. Higher latency, additional cost. Better for organization-wide discovery; overkill for single-agent local search.

Software development agency

Build on semble with DEV.co software developers

Cut LLM token consumption by 98% and enable coding agents to explore unfamiliar codebases instantly. Try Semble free—no GPU, no API keys, all local. Get started in under a minute with any MCP-compatible agent.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

semble FAQ

Does Semble require GPU or API keys?
No. Semble runs entirely on CPU with zero external dependencies. No API keys, cloud services, or GPU required. Indexes and searches are local.
How much faster is Semble than grep+read?
Indexing is ~200x faster than code-specialized transformers, queries are ~10x faster, and token consumption is ~98% lower than naive grep-and-read workflows. NDCG@10 is 0.854 (99% of transformer quality).
Can I use Semble with my favorite agent (Claude Code, Cursor, Codex)?
Yes. Semble integrates via MCP protocol with any MCP-compatible agent. Auto-install is available for Claude Code, Cursor, Codex, and OpenCode; manual setup is possible for others.
Is my code data stored or sent anywhere?
No. Semble indexes and searches locally on your machine. Indexes are cached in your OS cache folder (configurable). No telemetry, no remote uploads.

Software development & web development with DEV.co

DEV.co helps companies turn open-source tools like semble into production software. Our software development services cover the full lifecycle — architecture, web development, integration, and maintenance — delivered by software developers and web developers who ship. Engage our software development agency to implement or customize it for your mcp servers stack.

Accelerate Your Agentic Workflows with Semantic Code Search

Cut LLM token consumption by 98% and enable coding agents to explore unfamiliar codebases instantly. Try Semble free—no GPU, no API keys, all local. Get started in under a minute with any MCP-compatible agent.