Skip to content

System Settings & Vector Embeddings ​

The Settings View (Settings tab) provides administrative controls for configuring vector search engines, identity and authentication providers, secret backends, custom tool specifications, and enterprise access control rules.


🧭 Settings Sub-Navigation Tabs ​

System Settings Overview

The Settings interface is structured into five operational tabs:

Settings TabPrimary Configuration Domain
🧠 Vector & SearchLocal ONNX runtime parameters, external OpenAI/Ollama embedding APIs, and SIMD RRF search tuning
🪪 Identity & AuthActive Directory LDAP bindings, OIDC proxy headers, and OpenIddict OAuth 2.0 provider
🔐 Secret ProvidersHashiCorp Vault KV v2 AppRole/Token, Windows DPAPI Registry, and RFC 8693 token exchange
📂 Prompts & ResourcesCustom JSON specification file manager, prompt templates, and virtual resource definitions
👥 Access ControlRole-based access policies (ALLOW/DENY) and external Windows SID / AD group mappings

🧠 Tab 1: Vector & Search Engine (GeneralTab) ​

Settings Vector and Semantic Search Options

The vector embedding engine powers the Meta-Mode tool discovery function (search_tools). It generates mathematical representations of natural language intents and compares them against tool catalog metadata:

Vector Embedding & Semantic Search Configuration

  • Model Architecture: Uses the embedded All-MiniLM-L6-v2 transformer model (384-dimensional dense vectors).
  • Zero External Dependencies: Executes entirely in-process on CPU via Microsoft.ML.OnnxRuntime and Microsoft.ML.Tokenizers.
  • Data Privacy Guarantee: Tool queries and schema embeddings never leave the local container or host.
  • Automatic Provisioning: MCG downloads, validates SHA-256 checksums, and caches model weights in /data/models automatically.

2. External API Provider (OpenAI / Ollama / LiteLLM) ​

  • Usage: Offloads embedding generation to remote cloud APIs or self-hosted GPU inference clusters via IEmbeddingProvider.
  • Supported Backends: OpenAI (text-embedding-3-small, text-embedding-ada-002), Azure OpenAI, Ollama (/v1/embeddings), LiteLLM, Open WebUI, and vLLM.
  • SSRF Guardrails: MCG validates custom provider URLs (SecurityValidationHelper.IsPrivateOrLoopback) to prevent Server-Side Request Forgery.
  • Key Encryption: API keys are encrypted at rest in the database using AES-256-GCM envelope encryption.

3. In-Memory SIMD Vector Store & Reciprocal Rank Fusion (RRF) ​

  • In-Memory SIMD Acceleration (IToolVectorStore): Tool vectors are indexed in memory. Cosine similarities are calculated using .NET 10 hardware SIMD intrinsics (System.Numerics.Tensors.TensorPrimitives.CosineSimilarity), providing sub-millisecond query evaluation without requiring external vector databases for catalogs up to 5,000+ tools.
  • Reciprocal Rank Fusion (RRF, $k=60$): The search engine blends dense vector semantic similarity with multi-signal lexical matching across tool names, descriptions, tags, and JSON schemas.
  • Graceful Degradation: If embedding providers are disabled or offline, MCG seamlessly falls back to pure keyword ranking with zero downtime.

🪪 Tab 2: Identity & Auth Providers (IdentityAuthTab) ​

Settings Identity and Authentication Providers

Manage how MCG authenticates incoming client connections and extracts identity claims:

1. Active Directory / Windows SID Provider ​

  • Enable or disable native Windows Kerberos and NTLM token negotiation.
  • Configure Domain Controller hostnames, Base DN paths, and service account credentials.

2. OIDC / Reverse Proxy Headers Provider ​

  • Enable or disable forward-auth header extraction (Authentik, Authelia, PocketID, Keycloak, Traefik, Caddy, Nginx).
  • Configure custom header names: Remote-User, Remote-Groups, Remote-Email, and Remote-Name.

3. OpenIddict OAuth 2.0 Authorization Server ​

  • Configure token lifetimes, including access token validity and refresh token rotation policies.
  • Specify paths for X.509 signing certificates and encryption keys.

🔐 Tab 3: Secret Providers (SecretProvidersTab) ​

Settings Enterprise Secret Providers

Configure external credential vaults that MCG queries at runtime:

  • HashiCorp Vault KV v2: Configure Vault URL, AppRole credentials (roleId and secretId), or direct Vault Token with automatic renewal.
  • Windows Registry: Configure base keys (HKLM / HKCU) and paths for DPAPI-encrypted secrets.
  • Container Environment: Map host or container environment variables.
  • OAuth 2.0 Token Exchange (RFC 8693 / PocketID): Configure downstream token delegation and impersonation.

📂 Tab 4: Prompts & Resources File Manager (CustomFilesTab) ​

Settings Prompts and Resources File Manager

Create, edit, and manage custom JSON files that define virtual tools, prompt templates, and virtual resources:

  • File Catalog Table: Displays registered specification files by filename, type (Tools, Prompts, Resources), and last updated timestamp.
  • Interactive JSON Editor: Integrated Monaco/CodeMirror editor that validates JSON syntax before writing to disk or database.
  • Hot Reloading: Saving changes immediately updates in-memory caches and triggers semantic vector re-indexing.

👥 Tab 5: Access Control & Group Mappings (AccessControlTab) ​

Settings Access Control and Group Mappings

Configure fine-grained access rules and map external directory groups:

  • Group Mappings Table: Maps external directory groups (e.g. CN=IT-Admins,OU=Groups,DC=corp or S-1-5-21-1001) to internal MCG roles (full_admin, user).
  • Server Policies Table: Central policy management matrix defining target identifiers (server:docker, tool:docker__ps), required security groups, and evaluation modes (ALLOW or DENY).

Released under the Apache 2.0 License.