Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/affaan-m/ECC/llms.txt

Use this file to discover all available pages before exploring further.

Every tool description loaded by an active MCP server consumes tokens from your context window. A Claude Code session with 200k tokens and a dozen MCP servers enabled may effectively have only 70k tokens available for actual work. ECC treats this as a real engineering constraint: it ships a catalog of 33 ready-to-use MCP server definitions so you can pick the right subset for each project, copy them in, and keep everything else disabled.

​
What the MCP Catalog Provides

ECC’s MCP server definitions live in mcp-configs/mcp-servers.json. Each entry is a complete server config block — command, args, env vars, and type — ready to paste into your harness config. Placeholder values like YOUR_GITHUB_PAT_HERE mark fields that require real credentials before the server will work.
Keep under 10 MCP servers enabled per project and under 80 total tools active. Each MCP tool description eats context from your 200k window. Exceeding these limits degrades performance significantly — you may effectively only have ~70k tokens of usable context.
To add an MCP to Claude Code, copy the server definition into the mcpServers block of ~/.claude.json. To disable an MCP at runtime without removing it, use /mcp inside Claude Code; those choices persist in ~/.claude.json. For project-scoped MCP access, copy the desired entry from mcp-configs/mcp-servers.json into a project-local .mcp.json file.

​
Server Catalog by Category

​
Productivity & Project Management

jira

Jira issue tracking — search, create, update, comment, and transition issues. Uses uvx mcp-atlassian==0.21.0. Requires JIRA_URL, JIRA_EMAIL, and JIRA_API_TOKEN.

github

GitHub operations — PRs, issues, and repos. Uses npx @modelcontextprotocol/server-github. Requires GITHUB_PERSONAL_ACCESS_TOKEN.

confluence

Confluence Cloud integration — search pages, retrieve content, explore spaces. Uses npx confluence-mcp-server. Requires URL, email, and API token.

​
Memory & Context

memory

Basic persistent memory across sessions. Uses npx @modelcontextprotocol/server-memory. No credentials required.

omega-memory

Richer persistent agent memory with semantic search, multi-agent coordination, and knowledge graphs. Uses uvx omega-memory serve. Runs locally; no API key required.

longhand

Lossless Claude Code session history — indexes raw tool calls, file edits, and thinking blocks from ~/.claude/projects/*.jsonl into local SQLite + ChromaDB before Claude Code rotates them. Complements synthesized memory with verbatim recall. Install: pip install longhand && longhand setup.

squish

Local-first persistent memory runtime — auto-captures context across sessions, 1–20ms recall, 283KB footprint, SQLite storage. No second LLM required. Also available via squish run mcp.

memxus

Universal persistent memory across Claude Code, Cursor, Gemini CLI, and any AI tool — save context once, auto-recalled in every session. HTTP server at mcp.memxus.com. Requires MEMXUS_API_KEY. Free at memxus.com.

sequential-thinking

Chain-of-thought reasoning. Uses npx @modelcontextprotocol/server-sequential-thinking. No credentials required.

​
Web & Research

firecrawl

Web scraping and crawling. Uses npx firecrawl-mcp. Requires FIRECRAWL_API_KEY.

exa-web-search

Neural web search, research, and data ingestion. Uses npx exa-mcp-server. Requires EXA_API_KEY. Best used task-scoped for broader research after primary docs and GitHub search.

parallel-search

LLM-optimized web search and fetch — takes an objective plus queries and returns citation-backed excerpts in a single call. HTTP server at search.parallel.ai/mcp. Works key-free for anonymous use; add an Authorization header for higher rate limits.

context7

Live documentation lookup for libraries and frameworks. Uses npx @upstash/context7-mcp@latest. No credentials required. Use with the /docs command and documentation-lookup skill.

browser-use

AI browser agent for web tasks. HTTP server at api.browser-use.com/mcp. Requires x-browser-use-api-key header.

browserbase

Cloud browser sessions. Uses npx @browserbasehq/mcp-server-browserbase. Requires BROWSERBASE_API_KEY.

playwright

Browser automation and testing via Playwright. Uses npx @playwright/mcp --browser chrome. No credentials required.

​
Databases & Analytics

supabase

Supabase database operations. Uses npx @supabase/mcp-server-supabase@latest --project-ref=YOUR_PROJECT_REF. Replace YOUR_PROJECT_REF with your actual Supabase project reference.

clickhouse

ClickHouse analytics queries. HTTP server at mcp.clickhouse.cloud/mcp. No credentials required for cloud endpoint.

​
Deployments

vercel

Vercel deployments and projects. HTTP server at mcp.vercel.com. No credentials required.

railway

Railway deployments. Uses npx @railway/mcp-server. No credentials required.

cloudflare-docs

Cloudflare documentation search. HTTP server at docs.mcp.cloudflare.com/mcp.

cloudflare-workers-bindings

Cloudflare Workers bindings. HTTP server at bindings.mcp.cloudflare.com/mcp.

cloudflare-workers-builds

Cloudflare Workers builds. HTTP server at builds.mcp.cloudflare.com/mcp.

cloudflare-observability

Cloudflare observability and logs. HTTP server at observability.mcp.cloudflare.com/mcp.

​
AI Generation

fal-ai

AI image, video, and audio generation via fal.ai models. Uses npx fal-ai-mcp-server. Requires FAL_KEY.

​
Routing & Optimization

nexus

Local cost/privacy proxy — query your own usage and savings, route to the cheapest capable model, and mask secrets/PII before egress. Provides nexus_stats, nexus_savings, nexus_recent, nexus_providers, and nexus_cost_breakdown tools. Runs locally via nexus mcp command.

token-optimizer

Token optimization for 95%+ context reduction via content deduplication and compression. Uses npx token-optimizer-mcp.

​
ECC-Specific Servers

devfleet

Multi-agent orchestration — dispatch parallel Claude Code agents in isolated worktrees, auto-chain missions, and read structured reports. Local HTTP server at localhost:18801/mcp. See github.com/LEC-AI/claude-devfleet.

evalview

AI agent regression testing — snapshot behavior, detect regressions in tool calls and output quality. 8 tools including create_test, run_snapshot, run_check, and generate_visual_report. Uses python3 -m evalview mcp serve. Install: pip install "evalview>=0.5,<1".

codescene

Optional CodeScene Code Health MCP — structural code review and commit/PR gates. Uses npx @codescene/codehealth-mcp. Requires CS_ACCESS_TOKEN. Not enabled by default — opt-in only. See skills/codehealth-mcp/SKILL.md for data boundaries and failure behavior.

​
Framework-Specific

magic

Magic UI component library. Uses npx @magicuidesign/mcp@latest. No credentials required. Pair with the frontend-patterns module.

laraplugins

Laravel plugin discovery — search packages by keyword, health score, and Laravel/PHP version compatibility. HTTP server at laraplugins.io/mcp/plugins. Use with the laravel-plugin-discovery skill.

​
Filesystem

filesystem

Filesystem operations scoped to a specified directory. Uses npx @modelcontextprotocol/server-filesystem /path/to/your/projects. Set the path to the directory you want the agent to access.

​
Example Server Definitions

These are the exact entries from mcp-configs/mcp-servers.json. Copy them into your ~/.claude.json mcpServers block: 
{
  "nexus": {
    "command": "nexus",
    "args": ["mcp"],
    "description": "Local cost/privacy proxy - query your own usage & savings, route to the cheapest capable model, and mask secrets/PII before egress"
  }
}

​
Disabling ECC-Bundled MCPs

All ECC MCP entries are catalog-only — none are enabled by default. If you already run your own copies of ECC-bundled MCPs, set ECC_DISABLED_MCPS to prevent ECC’s install and sync flows from adding duplicates: 
export ECC_DISABLED_MCPS="context7,github"
ECC_DISABLED_MCPS is an ECC install/sync filter — it prevents ECC from re-adding those servers during install. It is not a live Claude Code toggle. To disable an already-loaded MCP at runtime, use /mcp inside Claude Code; those choices persist in ~/.claude.json.

Build docs developers (and LLMs) love

Get started for freeTalk to us