Twira

MCPAI代理

提供方 TwiraHQ

AI代理 未标注 API Key 社区 / 第三方 ClaudeCursorGemini CLICodex
查看接入方式

Coding power tools for AI agents. 18 deterministic MCP tools: code-graph search, SAST, blast radius, wire-level PII redaction, signed audit chain. Works with Claude Code, Codex, Gemini, Cursor. Free tier on install.

一句话选型结论

按资料字段整理

Twira 属于AI代理类 MCP 服务器,传输方式未标注,API Key,可直接接入 Claude、Cursor、Gemini CLI 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类AI代理 传输方式未标注 鉴权API Key 部署方式未标注
GitHub 星标 未标注 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

AI代理
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

code intelligence code search sast security compliance audit database code review

支持客户端

适用场景

code intelligencecode searchsast

安装与配置

未标注 API Key
上游未给出传输方式
brew install twirahq/tap/twira\r

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    按官方文档确认运行环境(常见为 Node.js 或 Python),装好后再执行安装命令。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
给代理接外部能力

让代理在任务中途调用搜索、数据库等真实工具。

多步任务编排

把一串操作拆成步骤,由代理按顺序执行。

自动化重复流程

把固定流程交给代理跑,减少手工操作。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude、Cursor、Gemini CLI,换客户端不用重新研究配置。

先想清楚的地方

  • 使用前要准备好API Key对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在AI代理场景里让模型直接调用外部能力的人。
  • 已经在用 Claude、Cursor 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Twira当前 未标注 未标注 未标注 有 未标注
Filesystem 560c4363 13 未标注 本地部署 无 2026-05-17
Server Everything (everything) 11 stdio · 本地进程 本地部署 有 2026-05-17
Time 7365cb6d 2 SSE · 服务端推送 远程部署 有 2026-05-18
Everything Claude Code 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Openmemory 557e1585 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

AI代理
字段来自项目仓库与公开索引,中文说明由本站标注
提供方TwiraHQ上游标注的维护方 TwiraHQ
分类AI代理本站按上游分类展示
连接方式未标注上游未给出传输方式
鉴权方式API Key远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · Cross-platform本地安装时用到的包标识与运行时
部署类型未标注上游给出的部署形态
最近核验未标注上游最后核验时间

详细介绍

7.7 KB
来自项目仓库与上游目录的原文,未做删改

Twira

Coding power tools for AI agents. Deliver better code, faster and safer.

18 PowerTools · 26 languages · 65 deterministic detectors ·
Wire-level PII redaction · Tamper-evident audit chain ·
Works with every MCP agent

Install ·
What is Twira ·
AI Compliance Proxy ·
The toolbelt ·
Free vs Pro ·
Get started ·
twira.com


Install```sh

curl (macOS / Linux)

curl -fsSL https://twira.com/install.sh | sh

npm

npm install -g @twira/cli

Homebrew (macOS / Linux)

brew install twirahq/tap/twira


One binary, no runtime dependencies, on macOS, Linux and Windows. No signup,
no account, no card.

## What is Twira?

Twira is a single local binary that gives your AI coding agent deterministic,
indexed access to your codebase, so it reads real code instead of guessing.
Larger context windows didn't fix the hallucinations; deterministic tools did.
Your agent is the operator. Twira is the power tool beneath it. You stay in
control.

Indexed search across 26 languages. 65 deterministic detectors. Tamper-evident
audit chain. Works with Claude Code, Codex, Gemini, Cursor, and anything else
that speaks MCP.

Every tool is reachable two ways: your AI agent calls them over MCP, and you
call the same tools from your terminal. Deterministic, local-first,
cryptographic where it matters. Your source code never leaves your machine.```json
{
  "mcpServers": {
    "Twira": {
      "command": "twira",
      "args": ["mcp"]
    }
  }
}

You never need to write that config by hand: twira init detects your agent
and wires it for you.

The AI Compliance Proxy

The flagship. A wire-level proxy that sits between your AI agents and the
model providers they call, on your machine, so personal data and secrets are
redacted before they leave it:```
your AI agent ──▶ Twira proxy ─────────────────▶ model provider
│ redact PII + secrets sees placeholder
│ swap in the custodied API key tokens, never the
│ sign + chain every call real values
you see normal ◀── restore real values ◀────────── response
output


- **Redaction at the wire.** Around 50 text patterns plus deterministic
  person-name detection, structured Article 9 identifiers (health, politics,
  religion), API keys and secrets, and OCR-driven redaction inside images:
  faces blurred, personal text blacked out, EXIF stripped.
- **Reversible, invisibly.** Each value becomes a session-scoped token on the
  way out and is swapped back on the way home. Your workflow never notices;
  the provider never sees the real data.
- **Key custody.** Provider API keys are stored AES-256-GCM encrypted and
  injected at the wire. The agent never sees the key. Rotate centrally
  without touching every machine.
- **Signed receipts.** Every call is Ed25519-signed and Merkle-chained,
  verifiable offline. Per-session evidence in the dashboard shows exactly
  what was redacted, what token replaced it, and how often it was sent, and
  exports as a signed redaction certificate.
- **Compliance postures.** Hospital, Bank and Government floors enforce
  Strict; General maps to Standard; Dev to Lenient. GDPR purge removes
  mappings and bodies on request while keeping the chain proof intact.
- **Spend visibility.** Per-agent token usage across every provider, in one
  place.

Built for the rules teams actually face: GDPR and UK GDPR, the EU AI Act,
ISO/IEC 42001, the NIST AI Risk Management Framework, Singapore's Agentic AI
framework, and equivalent regimes worldwide.

## The toolbelt

Eighteen PowerTools. Each one a single command, deterministic and local.

### Look up code

| Tool | What it does |
|---|---|
| **Index** | Your codebase as a queryable knowledge graph: symbols, call graph, dependencies, references and optional embeddings across 26 languages, kept fresh on every commit. |
| **Code Search** | Find anything fast. Five modes in one interface: symbol, path, content, regex and semantic (vector + keyword + call-graph, fused). |
| **Code Read** | A symbol slice, a file overview, or the whole file, without burning tokens re-reading what the index already knows. |
| **Impact** | Know what would break before you ship: references, dependency direction and blast radius from the call graph, risk-rated. |
| **Database MCP** | Code and database on one interface. Maps every table, FK, index, view and RLS policy live across 6 engines; read-only queries; finds every place code touches a table across 17 ORM patterns. |

### Find bugs and risks

| Tool | What it does |
|---|---|
| **Diagnose (SAST)** | 65 deterministic detectors across 4 profiles, locally, in the millisecond range. Baselines, suppressions that survive renames and refactors, output as JSON or SARIF 2.1.0. |
| **Dependency Vulnerabilities (SCA)** | OSV-backed and reachability-filtered (installed AND imported, so the noise drops). Local cache means air-gapped runs still work. |
| **Risk** | Triage what changed: RED, YELLOW, GREEN per commit, at a glance. |

### Data protection and evidence

| Tool | What it does |
|---|---|
| **AI Compliance Proxy** | The wire-level redaction, custody and receipts engine described above. |
| **Audit** | A tamper-evident, cryptographically signed, append-only record of every meaningful action the agent takes. RFC 3161 time-stamped, verifiable offline. |

### Coordinate the work

| Tool | What it does |
|---|---|
| **Team** | Ask, review, brainstorm and debate across 10+ frontier models from 6+ providers, synthesised into one peer-reviewed answer. |
| **Code Review** | Type `/code-review` in your agent: multiple frontier models review the commit, each with a different lens. |
| **Plan Review** | Type `/plan-review`: multiple models review the implementation plan before any code gets written. |
| **Masterplan** | One shared task graph that every agent in every session works from. Atomic claims, no duplicated work. |
| **Relay** | Parallel agent sessions on the same repo without collisions, coordinated by file claims. |
| **Lore** | Institutional memory across agents: save the lesson once, and every future session checks it before touching the file. |
| **Port** | Cross-language migration with structural matching. Port a 200,000-line legacy codebase without losing a function. |

### Defensibility and ergonomics

| Tool | What it does |
|---|---|
| **Localhost Dashboard** | The visual control panel on 127.0.0.1: redaction evidence, sessions, spend, audit, toggles, instructions. |
| **Notifications** | Desktop toasts, a chime, optional spoken alerts when an agent finishes, asks, or needs permission. |

## Free vs Pro

**Free, for ever, personal use.** Index, Code Search and Code Read. No
signup, no email, no card.

**Pro, the full toolbelt.** $29.99/month with a 14-day trial, no card
required. Free for students who verify with an institutional email. Full
comparison at twira.com/pricing.

## Works with

26 languages · 6 database engines · 6+ AI providers · Claude Code, Codex,
Gemini CLI, Cursor and any MCP-compatible agent · macOS / Linux / Windows ·
air-gap capable.

## Get started```sh
twira init       # set up Twira in your repo: wires your AI agent (MCP) and builds the index
twira login      # link this machine to your Twira account (Pro and trials)
twira dashboard  # open the local dashboard in your browser

From there your AI agent does the work. Ask it to search, read, check impact,
or diagnose, and it reaches for Twira's tools by itself.

Links

  • Website: https://twira.com
  • Pricing: https://twira.com/pricing
  • Contact: https://twira.com/contact
  • Releases: https://github.com/TwiraHQ/twira/releases

Free for personal use, straight from install. Pro unlocks the full toolbelt
with a 14-day trial, no card required.

常见问题

7 问
答案由本站依据条目字段整理,不引用上游原文
Twira 怎么安装?

在客户端按下面的安装命令接入:brew install twirahq/tap/twira\r。装完在客户端配置里引用同一个包或端点。

Twira 怎么在 Claude、Cursor、Gemini CLI 里配置?

它的传输方式未标注,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Twira 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Twira 需要 API Key 吗?

资料标注需要API Key认证,使用前准备好对应凭据,并按最小权限配置。

Twira 支持哪些客户端?

资料标注支持 Claude、Cursor、Gemini CLI、Codex、Zed;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Twira 能提供哪些能力?

资料显示能力标签包括 code intelligence、code search、sast、security、compliance、audit。具体工具清单以官方文档为准。

Twira 需要什么运行环境?

运行时为 Cross-platform。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 未标注。