Meta-MCP Command Center Capability-first · Local ownership · Capabilities, not servers
Meta-MCP Operating Layer

Capability-Centric MCP Operating Layer

Meta-MCP is your local-first hypervisor and super-server for normalized capabilities. Acquire locally, analyze runtime, introspect tools, register professionally, and normalize everywhere.

Local-first Maximum Local Ownership Normalized Capabilities Capabilities, not servers Raw tools stay behind the hypervisor
Mission Control

Autopilot MCP Import → Normalized Capability

The intended operator flow is one address only. The Hypervisor acquires the source, analyzes runtime, prepares dependencies, introspects MCP tools, registers raw tools, then maps them into our stable capability catalog.

01Acquire locally

Clone, mirror, or snapshot the MCP server under Artifact Store.

02Analyze runtime

Detect Node, Python, uv, Docker, stdio, HTTP, package scripts, credentials needs, and missing dependencies.

03Introspect tools

Run initialize + tools/list and capture the real schema returned by the server.

04Register professionally

Record raw tools with quality, protocol, runtime candidate, impact, and invocation evidence.

05Normalize as ours

Map raw upstream tools into stable names like meta.system.run_command.

Autopilot import & register

Paste one MCP server URL or Git repo. The system will do the rest.

The system will infer source type, clone or mirror locally, analyze runtime, prepare an import job, run introspection, register tools, rebuild indexes, and map results into normalized capabilities where possible.

URL → Local Ownership → Tools → Capabilities
Waiting for MCP server address.
Professional import is not complete at clone. Success requires: source acquired → repo analyzed → runtime ready → introspected → tools registered → normalized capability mapped. If tools are zero, the UI must explain the exact blocker.

Capability Catalog

Normalized, model-facing capabilities ready for controlled operator and model use.

Meta-MCP Server

The model-facing interface your agents should use first.

Latest Import Result

Live Activity Console

Recent import and invocation signals from the hypervisor.


          

Server Fleet & Quality

Imported MCP servers, runtime health, and readiness signals.

AI routing matrix

How a model should find the right tool

The model should never browse thousands of raw tools directly. It should search normalized capabilities, inspect exact use guidance, then invoke one stable capability ID.

NeedRun a host commandmeta.system.run_command
NeedSearch imported notesmeta.notes.search
NeedPerform arithmeticmeta.math.*
NeedInspect backend detailsdescribe_capability

Selected Capability / AI Usage Guidance

AI Model Tool Selection Contract

This is the intended flow for future models: discover categories first, inspect normalized capabilities, then invoke through our stable interface.

1search_capabilitiesFind the right normalized capability.
2describe_capabilityRead purpose, arguments, impact, and examples.
3invoke_capabilityRun our stable capability ID, not a raw upstream tool.
4quality feedbackLedger, graph, and quality scores improve routing.

Overview

Meta-MCP Hypervisor / Local-first / Maximum Local Ownership

Product North Star

Capabilities, not servers

Raw MCP sources are inputs. The hypervisor owns local state, normalizes tools into stable capabilities, and routes through a deterministic control surface.

North starManage capabilities instead of juggling upstream servers.Import → Normalize → Route → Observe
TrustKeep local_owned, local_mirrored, remote_only, and quarantined visible everywhere.Show ownership in the UI
RoutingPrefer deterministic, explainable ranking before semantic automation.Explain every selection
Golden pathImport, introspect, register, test invoke, then rebuild ledger and graph.One operator flow

Recent Invocations

Recent Errors

Import MCP Server

Create a manifest, run ingestion, inspect capabilities, and register tools.

Import Summary

Readiness at a glance

The active import form is translated into a small operator summary so blockers and next steps stay visible before you run the pipeline.

One-field import

Automatic MCP Import Wizard

Paste only the MCP server URL or local path. The dashboard will infer the manifest, create the import job, run the pipeline, show live progress, and surface registered tools or exact failure reasons.

No manual fields required first
Ready. Paste a link to begin.
Autopilot contractPaste URL only. Manual fields remain as override, not the starting point.
Tool registration contractRegistered tools must include real schema, transport, runtime mode, quality, and normalized mapping status.
Model contractEvery successful import should explain what the server does in language useful for an AI model.

Latest Import Job & Result

Step 1: Source

Step 2: Runtime

Step 3: Actions

Remote endpoint registration is recognized but execution remains controlled by current policy.

Step 4: Progress Timeline

Server Fleet

External MCP Quality

Raw Tool Registry

Diagnostics layer for imported upstream tools. Model-facing usage should prefer normalized capabilities.

Saved Views

Common operator filters

Jump directly to high-signal subsets instead of retyping the same filters every time.

Tool Details

Runtime Monitor

Supervised Processes

Runtime Units

Host Runtime

Check host dependencies, prepare local repositories, and retry MCP introspection.

Host Dependencies

Runtime Missing

Select an artifact or use the default mcp-server-commands plan.

        

Ledger

Summary

Recent Invocations

Recent Errors

Graph

Graph Summary

Categories

Hot Tools

Error Tools

Normalized Capabilities

Our own stable capability names mapped from raw imported MCP tools.

Normalized Summary

Capabilities

Selected Capability


        

Meta-MCP Super Server

Model-facing MCP endpoint over normalized capabilities, not raw upstream tool names.

Model-facing contract

One MCP server. Five stable tools. Unlimited normalized capabilities.

The LLM should use these meta-tools first, then invoke our normalized capability IDs. Raw upstream tools remain available behind the hypervisor for execution and diagnostics only.

search_capabilitiesFind the right owned capability by intent.
describe_capabilityRead precise purpose, inputs, impact, and backend quality.
invoke_capabilityExecute a stable capability like meta.system.run_command.
list_categoriesExplore our taxonomy instead of raw tool names.
explain_serverSummarize imported MCP servers and what they provide.

Server Summary

Model-Facing Tools

Capability Search


        

Invoke Capability


        

Invocation Audit

Recent model-facing capability invocations, policy decisions, and raw routing targets.

AI Tool Selection Guide

Policy

Profile Details

Tool Override

System Actions

Ready Select an action to run a healthcheck, rebuild, or cleanup task.

Meta-MCP Service

Proxy Status

Reserved Ports

Public URLs


            

Deployment Profile


            

Action Output

Release / Backup

Latest Release

Release Paths


            

Manifest Summary


            

Release Output

Invoke Test

Tool Details

{}

Command Helpers


      

Import New MCP Server

Paste a URL or path, generate a manifest, run the pipeline, inspect files, register tools, and test invoke.

Paste-only mode

Autopilot MCP Server Import

Enter a GitHub URL, local folder, or endpoint. The wizard will infer artifact ID, name, source type, transport, runtime hints, then run import and tool registration with live progress.

Auto-detect → Analyze → Register
Paste an address, then use Auto-fill or Run Full Autopilot Import.

Repository Analysis

not analyzed
Paste a Git URL or local path, then click Analyze Repository.
Local ownership rule: external URLs are source acquisition inputs only. The Hypervisor clones, analyzes, and extracts locally; the AI model should only use normalized Meta-MCP capabilities, not upstream raw tools directly.

Prepare Commands


                

Evidence


                
Runtime Missing panel will show missing host dependencies after analysis.

Runtime Candidates

untested

              
Secrets injection will be handled by a later phase.

Generated Ingestion Manifest Preview


          

          

Live Import Progress

active
0%
Waiting for import activity...

Stages

Runtime Activity

Live Activity Console

Job ID: none | Lines: 0 | Auto-scroll: on | Last poll: never

            
Import results will appear here.
Why no tools were registered? Check source acquisition, runtime readiness, credentials, and whether the backend actually exposes tools.

Imported Files

Registered Tools

No tools registered yet. Inspect the runtime plan, host dependencies, credentials, and the "Why no tools were registered?" explanation.