Overview

/mcp is a production command in the Infrastructure category of the Prismatic Platform. It provides complete Model Context Protocol operations and management, serving as the unified control interface for all MCP-related activities across the platform. The Model Context Protocol is the standard communication layer that enables AI agents to interact with external tools, data sources, and services through a structured, capability-negotiated interface.

The Prismatic Platform's MCP infrastructure encompasses 14 or more MCP servers, collectively providing over 100 tools spanning file system operations, version control, database queries, semantic search, memory management, and platform-specific capabilities. The /mcp command provides a single point of control for this entire infrastructure: server management, connection monitoring, tool discovery, capability enumeration, performance analysis, and configuration management.

This command operates under the L2+ authority level and is executed by the mcp-evolution-coordinator agent, the platform's primary MCP lifecycle manager. It is part of the platform's 216-command slash command registry, built on the AIAD (Autonomous Intelligence Agent Design) standard. The medium usage frequency reflects its role as both a routine operational tool (status checks, connection management) and an occasional administrative tool (configuration changes, server management).

The command's design follows the principle of progressive disclosure: simple subcommands for common operations (status, connect, disconnect) and more detailed subcommands for administrative tasks (configure, diagnose, benchmark). This layered interface ensures that routine operations remain quick and accessible while advanced functionality is available when needed.

Architecture

The /mcp command acts as a facade over the platform's MCP management infrastructure, coordinating multiple subsystems through a unified command interface.

+---------------------+     +----------------------+     +-------------------+
|  Command Router     |---->|  Server Manager      |---->|  Connection Pool  |
|  (Subcommand Parse) |     |  (Lifecycle Ctrl)    |     |  (Channel Mgmt)   |
+---------------------+     +----------------------+     +-------------------+
         |                           |                           |
         v                           v                           v
+---------------------+     +----------------------+     +-------------------+
|  Tool Registry      |     |  Config Manager      |     |  Health Monitor   |
|  (Capability Map)   |     |  (Server Profiles)   |     |  (Liveness Check) |
+---------------------+     +----------------------+     +-------------------+
         |                           |                           |
         v                           v                           v
+---------------------+     +----------------------+     +-------------------+
|  Performance Tracker|     |  Diagnostics Engine   |     |  Telemetry Sink   |
|  (Latency/Throughput)     |  (Error Analysis)    |     |  (Event Stream)   |
+---------------------+     +----------------------+     +-------------------+

The Server Manager controls the lifecycle of individual MCP servers: startup, shutdown, restart, and health monitoring. It maintains a registry of all configured servers with their current state, connection parameters, and capability advertisements.

The Connection Pool manages active communication channels between the platform and MCP servers. Each channel is a supervised OTP process that handles message serialization, deserialization, timeout management, and automatic reconnection. Channels support both stdio-based and TCP-based transport depending on the server configuration.

The Tool Registry maintains a unified index of all tools exposed by all connected MCP servers. Each tool entry includes its name, description, parameter schema, server affinity, and usage statistics. This registry enables the platform's agents to discover and invoke tools without needing to know which specific MCP server hosts them.

The Performance Tracker records latency, throughput, and error rates for every tool invocation, enabling operators to identify performance bottlenecks and degraded servers. Historical performance data supports trend analysis and capacity planning.

Usage

Status and Discovery

# Show overview of all MCP servers and their status
/mcp status

# List all available tools across all connected servers
/mcp tools

# Show detailed information about a specific tool
/mcp tool-info prismatic-mcp:search_code

# Discover tools matching a pattern
/mcp tools --search="file"

Connection Management

# Connect to a specific MCP server
/mcp connect prismatic-mcp

# Disconnect from a server
/mcp disconnect memory

# Reconnect a server (disconnect + connect)
/mcp reconnect github

# Connect to all configured servers
/mcp connect --all

Server Management

# Start a specific MCP server
/mcp start prismatic-mcp

# Stop a running server
/mcp stop memory

# Restart a server
/mcp restart github

# Start all configured servers
/mcp start --all

Performance and Diagnostics

# Show performance metrics for all servers
/mcp performance

# Run diagnostic health checks
/mcp diagnose

# Benchmark a specific tool
/mcp benchmark prismatic-mcp:search_code --iterations=100

# Show error analysis for a server
/mcp errors github --period=1h

Configuration

# Show current MCP configuration
/mcp config

# Update server configuration
/mcp config set prismatic-mcp.timeout 30s

# Export configuration for backup
/mcp config export --output=mcp-config.json

# Import configuration
/mcp config import mcp-config.json

Options & Parameters

Option	Type	Default	Description
`status`	Subcommand	--	Show MCP infrastructure overview
`tools`	Subcommand	--	List all available tools
`tool-info`	Subcommand	--	Show detailed tool information
`connect`	Subcommand	--	Connect to an MCP server
`disconnect`	Subcommand	--	Disconnect from an MCP server
`reconnect`	Subcommand	--	Reconnect to an MCP server
`start`	Subcommand	--	Start an MCP server process
`stop`	Subcommand	--	Stop an MCP server process
`restart`	Subcommand	--	Restart an MCP server
`performance`	Subcommand	--	Show performance metrics
`diagnose`	Subcommand	--	Run diagnostic checks
`benchmark`	Subcommand	--	Benchmark a specific tool
`config`	Subcommand	--	Configuration management
`errors`	Subcommand	--	Error analysis
`--all`	Flag	false	Apply operation to all servers
`--search`	String	none	Filter tools by name pattern
`--format`	String	table	Output format (table, json, markdown)
`--period`	Duration	24h	Time period for metrics and error analysis
`--iterations`	Integer	10	Number of iterations for benchmarks
`--verbose`	Flag	false	Show detailed output

Execution Flow

Subcommand Parsing -- The command router identifies the requested operation from the first positional argument and routes to the appropriate handler. Unknown subcommands receive a help message listing valid operations.
Server Resolution -- For operations targeting specific servers, the server name is resolved against the server registry. Unrecognized server names produce an error listing available servers.
Pre-condition Validation -- Operation-specific pre-conditions are checked. For example, connect verifies the server is started but not already connected; stop verifies the server is currently running.
Operation Execution -- The requested operation is performed through the appropriate subsystem (Server Manager, Connection Pool, Config Manager, etc.). Each operation is atomic with rollback on failure.
Result Verification -- Post-operation state is checked to confirm the operation achieved the desired effect. For lifecycle operations, this includes process state verification and health checks.
Output Rendering -- Results are formatted according to the --format parameter. The default table format provides a concise, aligned view suitable for terminal display.
Telemetry Recording -- All operations are recorded in the telemetry stream for audit trail and operational visibility.

Integration Points

Component	Integration Type	Description
Prismatic Agents	Agent Execution	Executed by the `mcp-evolution-coordinator` agent
MCP Protocol	Core Protocol	Central management interface for all MCP operations
Prismatic MCP	Primary Server	Platform's primary MCP server with 27+ native tools
Quality Gates	Pre/post validation	Quality checks on configuration and operational state
Telemetry	Observability	All operations emit structured telemetry events
Prismatic API	REST Exposure	MCP status and tool information available via REST API
Session Lifecycle	Auto-integration	MCP status verified during session initialization

Best Practices

Regular Status Checks: Run /mcp status at the beginning of each working session to verify all expected MCP servers are connected and healthy. Address any degraded servers before starting dependent work.

Tool Discovery Before Use: Use /mcp tools --search=<pattern> to discover available tools before writing automation that depends on specific tool capabilities. Tool availability may vary by environment.

Performance Monitoring: Periodically review /mcp performance to identify servers with degraded latency or elevated error rates. Address performance issues before they impact agent operations.

Configuration Version Control: Export MCP configuration (/mcp config export) and maintain it in version control alongside other platform configuration. This enables reproducible environments and configuration rollback.

Graceful Shutdown: Always use /mcp stop rather than killing MCP server processes directly. Graceful shutdown ensures in-flight requests complete and resources are properly released.

Error Handling

Error Condition	Response	Recovery
Server not found	Error listing available servers	Check server name spelling; verify configuration
Connection refused	Error with server address and port	Verify server is running; check network/firewall
Tool not found	Error listing similar tools	Use `/mcp tools` to discover available tools
Config validation failure	Detailed validation errors	Correct configuration values per error messages
Benchmark timeout	Partial results with timeout indicator	Increase timeout or reduce iteration count
Permission denied	Authority level requirement shown	Escalate to L2+ authority

Advanced Usage

Tool Invocation Testing

Test tool invocations directly through the MCP command for debugging and validation:

# Invoke a specific tool with test parameters
/mcp invoke prismatic-mcp:search_code --params='{"query":"GenServer","limit":5}'

# Test tool with timing
/mcp invoke --timed filesystem:read_file --params='{"path":"/tmp/test.txt"}'

Server Discovery

Discover MCP servers available in the environment but not yet configured:

# Scan for available MCP servers
/mcp discover

# Auto-configure discovered servers
/mcp discover --auto-configure

Bulk Operations

Execute operations across multiple servers efficiently:

# Restart all servers in a specific category
/mcp restart --category=development

# Health check with timeout override
/mcp diagnose --all --timeout=5s

Doctrine Compliance

All commands operate under the NO MERCY, NO DOUBTS doctrine:

NO MERCY: Zero tolerance for unreported failures. Every MCP operation must produce a clear success or failure outcome with actionable details. Connection drops, tool invocation failures, and configuration errors are all surfaced immediately with diagnostic context.
NO DOUBTS: Full state verification for all lifecycle operations. The command never assumes operational success based on the absence of errors -- it actively verifies post-operation state through health probes and state queries. Performance metrics provide evidence-based assessment of system health.

/mcp-autoboot - Start MCP infrastructure with full orchestration
/mcp-service - Manage Prismatic MCP server as macOS service for persistent operation
/ollama - Local AI Ollama model management, installation and optimization
/gardener - GARDEN legacy knowledge repository management across 116 repos
/garden-explore - Explore GARDEN repositories for patterns and knowledge
/connect - MCP server connection management across 14+ servers

Connect & Contribute

Created by Tomáš Korcak (korczis) | Open Source under GHL

GitHub | GitLab | LinkedIn | Contact
Developer Portal | Architecture | Meet the Creator

/mcp