@metago-ai/algorithms — 927 算法 MCP 服务器

MetaGO Agent Harness 的算法核心 — 57 工具 / 927 算法 / 14 触发大类 / 4 层架构，Engine V2.1.0 硬驱动。

What it is

@metago-ai/algorithms is the algorithmic backbone of the MetaGO Agent Harness. It exposes 57 MCP tools backed by 927 algorithms across 14 trigger categories, organized in a 4-layer architecture. Every tool is hard-driven by @metago-ai/[email protected] (KMWI memory + evolution engine + skill generator).

This is not a library of math functions. It is a Model Context Protocol (MCP) server that gives any MCP-compatible AI agent (Trae, Claude Code, Codex, Cursor, etc.) real-time access to coupling calculation, value assessment, distance metrics, correlation analysis, dimensionality reduction, security audit, frequency processing, and 7 more trigger categories.

Architecture — 4 layers / 14 trigger categories

| Layer | Name | Tools | Algorithms | Trigger Categories | |------|------|-------|------------|-------------------| | L1 | 通用入口层 (General Entry) | 3 | 9 | 3 | | L2 | 检索发现层 (Retrieval & Discovery) | 4 | 18 | 4 | | L3 | 高频专用层 (High-Frequency Specialized) | 30 | 540 | 4 | | L4 | 族级工具层 (Family Tools) | 20 | 360 | 3 | | Total | — | 57 | 927 | 14 |

The 14 trigger categories

| # | Category | Layer | Trigger condition | |---|----------|-------|------------------| | 1 | Algorithm execution | L1 | Agent needs to execute an algorithm | | 2 | Algorithm retrieval | L1 | Agent needs to find an algorithm by name/family/capability | | 3 | Statistics | L1 | Agent needs server statistics | | 4 | Family listing | L2 | Agent needs to list algorithm families | | 5 | Algorithm search | L2 | Agent needs semantic search over algorithms | | 6 | Info lookup | L2 | Agent needs detailed info on a specific algorithm | | 7 | Proactive suggestion | L2 | Agent needs algorithm recommendations for a task | | 8 | Coupling calculation | L3 | Task involves relationship strength / coupling degree | | 9 | Value assessment | L3 | Task involves multi-dimensional value scoring | | 10 | Distance metrics | L3 | Task involves similarity / distance between vectors | | 11 | Correlation analysis | L3 | Task involves statistical correlation | | 12 | Dimensionality reduction | L3 | Task involves PCA / SVD / feature reduction | | 13 | Security audit | L3 | Task involves vulnerability / security assessment | | 14 | Frequency processing | L3 | Task involves frequency-domain analysis |

Install

npm install @metago-ai/algorithms

Use as MCP server

Trae / Claude Code / Cursor / any MCP client

Add to your MCP client config:

{
  "mcpServers": {
    "metago-algorithms": {
      "command": "node",
      "args": ["node_modules/@metago-ai/algorithms/index.js"]
    }
  }
}

Standalone

npx @metago-ai/algorithms
# or
node node_modules/@metago-ai/algorithms/index.js

Tool list (57 tools)

L1 — General Entry (3 tools)

• execute — Execute any algorithm by name with parameters
• execute_sync — Synchronous algorithm execution
• get_statistics — Server statistics (tool count, algorithm count, layer distribution)

L2 — Retrieval & Discovery (4 tools)

• list_algorithms — List all 927 algorithms with metadata
• list_families — List 14 algorithm families
• search_algorithms — Semantic search over algorithm descriptions
• get_algorithm_info — Detailed info on a specific algorithm

L3 — High-Frequency Specialized (30 tools)

Coupling (8 tools)

• coupling_calculate · coupling_clustering · coupling_trend · normalize_coupling
• build_coupling_matrix · sort_coupling_scores · is_superconductive · cooccurrence_frequency

Value Assessment (3 tools)

• value_assess · build_value_vector · evaluate_bidirectional

Distance Metrics (6 tools)

• euclidean_distance · manhattan_distance · chebyshev_distance
• cosine_similarity · dice_coefficient · jaccard_coefficient

Correlation (4 tools)

• pearson_correlation · spearman_correlation · kendall_tau · covariance

Dimensionality Reduction (3 tools)

• extract_pca · simplified_svd · reduce_dimension

Security & Audit (3 tools)

• security_assess · audit_check · bias_detect

Frequency & Time (3 tools)

• frequency_process · time_analyze · time_decay_cooccurrence

L4 — Family Tools (20 tools)

Fuzzy & Semantic (3 tools)

• fuzzy_string_match · semantic_similarity · weighted_cosine

Reasoning (3 tools)

• reasoning_chain · logic_reason · conflict_resolve

Asymmetry & Symmetry (3 tools)

• detect_asymmetry · record_symmetric · identify_strong_pairs · identify_weak_pairs

Negentropy (1 tool)

• negentropy_measure

Intuition (1 tool)

• intuition_gauge

Lifeform Operations (8 tools)

• memory_operate · learning_learn · creation_create · evolution_evolve
• decision_decide · dialog_manage · idea_generate · proactive_suggest

Engine dependency

This package is hard-driven by @metago-ai/[email protected]:

• KMWI Memory — 4-layer memory (Knowledge → Memory → Wisdom → Intuition) backs algorithm state
• EvolutionEngine — 5-stage meta-evolution can generate new algorithms on demand
• DecisionLock — 4-gate verification before every algorithm output
• SkillGenerator — generates new algorithm SKILL.md files from internal patterns

License

MIT © 2026 易霄 / 元构光年（成都）人工智能科技有限公司

Links

• GitHub
• npm
• Homepage
• Issues