← 返回
未分类 Key

Nella

Provides codebase-aware search and indexing for non-trivial repos, enabling symbol location, call tracing, dependency checks, and persistent session context.
为复杂仓库提供代码感知搜索与索引,支持符号定位、调用追踪、依赖检查和持久会话上下文。
pablomanjarres
未分类 clawhub v0.2.7 1 版本 99595.1 Key: 需要
★ 0
Stars
📥 246
下载
💾 0
安装
1
版本
#latest

概述

Nella MCP

Nella is a codebase intelligence layer for AI coding agents. It exposes the repository as a set of searchable, structured tools over the Model Context Protocol, so an agent can ground its reasoning in the actual code rather than guessing from filenames or partial snippets.

When to reach for Nella

Use the Nella tools when any of the following is true:

  • The repo is non-trivial (more than a handful of files) and a grep would return too many false positives.
  • The user asks about behavior, call sites, or relationships rather than literal strings.
  • The session involves multiple turns and context should persist (assumptions, prior searches, prior decisions).
  • A change is being scoped and dependency drift or impact analysis matters.

If the question is purely lexical (find a literal token in one file), plain grep or view is faster and Nella is overkill.

Setup

The package ships two binaries: nella (CLI) and mcp (stdio MCP entrypoint).

Local stdio config for an MCP client:

{
  "mcpServers": {
    "nella": {
      "command": "npx",
      "args": ["-y", "@getnella/mcp", "--workspace", "/absolute/path/to/project"]
    }
  }
}

Hosted config (recommended for shared or always-on workspaces):

{
  "mcpServers": {
    "nella": {
      "url": "https://mcp.getnella.dev/mcp",
      "headers": { "Authorization": "Bearer nella_your_key_here" }
    }
  }
}

Quick shortcut for Claude Code: nella setup. For other clients: nella connect --client .

Available tools

ToolWhen to call it
-----------------------
nella_indexFirst contact with a workspace, or after large refactors. Pass --force to rebuild from scratch.
nella_searchDefault lookup. Supports hybrid (best general default), semantic (concept-level), and lexical (exact tokens).
nella_get_contextPull the current session memory before answering, so prior assumptions and searches are not lost.
nella_add_assumptionRecord any non-trivial assumption ("this function is only called from the worker") so it can be verified later.
nella_check_assumptionsReview recorded assumptions, especially before committing changes that depend on them.
nella_check_dependenciesDetect drift between the index and the working tree, or surface upstream impact of a change.
nella_heartbeatVerify trust-chain continuity between tool calls in long sessions.

Recommended flow

  1. Call nella_get_context at the start of a non-trivial task to load any prior session state.
  2. If the workspace has not been indexed yet, call nella_index. Skip this if the index is already current.
  3. Use nella_search (hybrid by default) to locate the relevant code before reading files.
  4. Record any load-bearing assumption with nella_add_assumption so it survives across turns.
  5. Before finalizing changes, run nella_check_assumptions and nella_check_dependencies.

Search mode selection

  • hybrid: Default. Combines lexical and semantic signals. Use when unsure.
  • semantic: Conceptual queries ("where do we handle rate limiting", "auth flow for OAuth"). Tolerates paraphrase.
  • lexical: Exact identifiers, error strings, or known token shapes. Faster, no embedding cost.

Notes for the agent

  • Always pass an absolute path to --workspace. Relative paths fail silently in some clients.
  • Hosted mode requires a valid API key. If nella_search returns an auth error, surface it to the user rather than retrying blindly.
  • The index can lag behind the working tree on large repos. If results look stale, call nella_index (without --force) for an incremental refresh.
  • Treat nella_get_context output as authoritative for prior decisions in the session. Do not override silently.

版本历史

共 1 个版本

  • v0.2.7 当前
    2026-05-21 14:49 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-intelligence

self-improving agent

pskoett
捕获经验教训、错误及修正内容,以实现持续改进。适用于以下场景:(1)命令或操作意外失败;(2)用户纠正Claude(如“不,那不对……”“实际上……”);(3)用户请求的功能不存在;(4)外部API或工具出现故障;(5)Claude发现自身
★ 4,073 📥 806,275
ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,374 📥 319,866
security-compliance

Skill Vetter

spclaudehome
AI智能体技能安全预审工具。安装ClawdHub、GitHub等来源技能前,检查风险信号、权限范围及可疑模式。
★ 1,223 📥 267,442