Model Context Protocol Configs
Relevant source files
The following files were used as context for generating this wiki page:
The Model Context Protocol (MCP) Configs system provides a structured framework for connecting AI models to external services, tools, and data sources. Within the Everything Claude Code (ECC) ecosystem, these configurations define how the orchestrator and specialized subagents interact with third-party platforms like GitHub, Supabase, and Vercel to retrieve real-time data and execute remote operations.
This system manages the lifecycle of MCP server connections, including registration, environment variable handling, and model-level tool exposure. By leveraging these configs, developers can extend the capabilities of AI agents beyond their training data, enabling actions such as direct SQL execution, repository management, and live documentation lookups.
Sources: the-shortform-guide.md:129-131, the-longform-guide.md:41-45, agents/docs-lookup.md:33-40
Architecture and Configuration Flow
The MCP system operates as a prompt-driven wrapper around external APIs. Configurations are typically stored at the user level or within specific project directories to control which tools are active during a session.
Configuration Hierarchy
MCP configurations are resolved using a specific precedence logic:
- User-Level Configs: Global settings typically found in
~/.claude/settings.json. - Project-Level Configs: Repository-specific overrides stored in
.mcp.json. - Default Catalog: Pre-defined configurations provided by the ECC project in
mcp-configs/mcp-servers.json.
Sources: the-shortform-guide.md:144-148, mcp-configs/mcp-servers.json
Integration Flow
The following diagram illustrates how the Orchestrator utilizes MCP configurations to bridge the gap between user requests and external services.
This diagram shows the decision-making process for invoking MCP-based tools versus standard local commands.
Sources: the-shortform-guide.md:133-140, commands/multi-workflow.md:30-40
MCP Server Management
Servers are defined in a JSON format specifying the transport method (STDIO or HTTP), execution commands, and necessary arguments.
Key MCP Server Definitions
| Server Name | Type | Command / URL | Purpose |
|---|---|---|---|
github | stdio | npx -y @modelcontextprotocol/server-github | Repository and PR management |
supabase | stdio | npx -y @supabase/mcp-server-supabase | Direct database and SQL execution |
context7 | http | https://mcp.context7.io | Live documentation and API reference |
vercel | http | https://mcp.vercel.com | Deployment and project status |
memory | stdio | npx -y @modelcontextprotocol/server-memory | Long-term knowledge graph storage |
Sources: the-shortform-guide.md:213-230, the-longform-guide.md:130-135, mcp-configs/mcp-servers.json
Context Window Optimization
A critical aspect of MCP configuration is the "Context Window Management" rule. Large numbers of enabled tools consume significant portions of the AI's token limit (context window). ECC guidelines recommend:
- Configuring many servers but keeping under 10 enabled at a time.
- Maintaining under 80 total active tools to prevent performance degradation.
- Disabling redundant MCPs that wrap existing CLI tools (e.g., using the GitHub CLI directly instead of the MCP).
Sources: the-shortform-guide.md:142-152, the-longform-guide.md:46-52
Documentation and Tooling Implementation
Specialized agents use MCP configurations to perform complex lookups. For instance, the docs-lookup agent relies on the context7 MCP server to resolve library IDs and query current documentation.
Sequence: Documentation Lookup via MCP
The workflow for fetching up-to-date API information using the Model Context Protocol.
Sources: agents/docs-lookup.md:42-65, the-shortform-guide.md:224
Implementation Example: MCP Config Entry
{
"mcpServers": {
"supabase": {
"command": "npx",
"args": [
"-y",
"@supabase/mcp-server-supabase@latest",
"--project-ref=YOUR_PROJECT_REF"
],
"env": {
"SUPABASE_SERVICE_ROLE_KEY": "YOUR_KEY"
}
}
}
}
A standard configuration entry used to initialize a database-focused MCP server.
Sources: the-shortform-guide.md:216-220, the-longform-guide.md:41-45
Resource and Tool Access Control
MCP configs determine tool naming conventions within the environment. Tools are often prefixed by the harness to avoid name collisions.
- Prefixing: Tools may appear as
mcp__[server_name]__[tool_name]. For example,mcp__context7__query-docs. - Security: Agents are instructed to treat fetched MCP content as untrusted data, applying sanitization before use.
- Routing: The Orchestrator routes tasks to specific MCPs based on the "Task Type" (e.g., Backend tasks to logic MCPs, Frontend tasks to design-focused MCPs).
Sources: agents/docs-lookup.md:40-45, commands/multi-workflow.md:110-120, the-shortform-guide.md:154-160
Summary
Model Context Protocol Configs are the essential plumbing for ECC's agentic workflows, enabling real-time interaction with the broader software ecosystem. By maintaining a modular and lean configuration, the system balances the need for powerful external tools with the constraints of the model's context window. Proper management of these configs ensures that agents can access live documentation, execute database queries, and manage cloud infrastructure while following project-specific security and style guidelines.