TengineAI
Run MCP tools in production without managing your own server — built-in retries, permissions, and observability.
Introduction
What is TengineAI?
TengineAI is execution infrastructure for AI tool calls.
It provides a remote MCP server that exposes your HTTP APIs as tools to AI runtimes. You define which endpoints the model can call. TengineAI handles authentication, execution, identity injection, and usage control — and returns structured results to the model.
TengineAI does not provide prebuilt third-party integrations. Instead, you define the HTTP APIs your model can call, and TengineAI handles the rest. Think of it as the execution layer between the model and your services.
TengineAI is to MCP tools what API Gateway is to REST APIs.
TengineAI works with any MCP-compatible model SDK or client. Anthropic is used throughout the documentation as the reference implementation because it is the most widely used and fully tested today.
Who Is This For?
TengineAI is built for developers who are building AI-powered products and need tool calls to execute safely against real APIs.
TengineAI is the right fit if:
- You are building a multi-tenant AI SaaS and need tool calls attributed to individual users
- You want to expose your own internal or external APIs to a model without building MCP tooling from scratch
- You need centralized control over which tools are available, with the ability to enable/disable them without redeploying
- You want usage visibility for tool calls across your project
- You want outbound requests to your APIs to be authenticated (HMAC-signed or bearer token) and carry member identity context
TengineAI is not the right fit if:
- You only need the model to call a single static endpoint — a direct API call in your code is simpler
- You want to execute tools entirely offline with no external network access
- You need the model to execute arbitrary code or manage infrastructure — TengineAI is tool execution, not an agent runtime
How TengineAI Works
┌─────────────┐ ┌──────────────────┐ ┌─────────────────────┐
│ │ │ │ │ │
│ MCP Client │────────▶│ TengineAI │────────▶│ Your APIs │
│ (Runtime) │ │ (Execution Layer)│ │ (CRM, billing, │
│ │◀────────│ │◀────────│ internal services, │
└─────────────┘ └──────────────────┘ │ public APIs) │
└─────────────────────┘
Copy
Flow:
- An MCP client (Claude Desktop, Cursor, or programmatic SDK) connects to TengineAI
- The client authenticates using an API key or a user-scoped session token
- TengineAI returns the project's enabled tools — only the ones you defined
- The model selects a tool to execute
- TengineAI renders the request template, injects member identity, signs the request (if configured), and forwards it to your API
- Your API executes the request and returns a response
- TengineAI returns the response to the model
- The model receives the result and continues execution
Core Concepts
TengineAI uses five primitives:
Projects
Isolated execution environments. Each project has its own API keys, integrations, and enabled tools. A token from one project cannot access another project's tools or configuration.
Tools
HTTP endpoints you register with TengineAI that the model can call. You define the name, description, HTTP method, URL template, input schema, and authentication strategy. TengineAI handles execution.
Tools are opt-in per project. The model only sees what you enable.
Integrations
MCP client connections. An integration defines how a model runtime (Claude Desktop, Cursor, custom SDK) authenticates with TengineAI.
API Keys
Project-scoped credentials (tengine_...). They authenticate requests from your application to TengineAI. Use them for server-to-server workflows and automation pipelines where all calls share a single project identity.
User-Scoped Sessions (Member Session Tokens)
Short-lived JWTs (tng_mst_..., 15-minute TTL) that scope an MCP session to a specific end user. Your backend mints them server-side by presenting a cryptographically signed member_assertion. TengineAI verifies it and issues the session token.
Use member session tokens when each end user should have their tool calls attributed to them individually.
Reliability Guarantees
- Retries won't duplicate writes — Safe retry handling prevents duplicate side effects when the model or SDK retries
- Tool calls are logged — Every invocation has a Run ID and appears in Run history
- Keys revoke across sessions — Revocation takes effect immediately for all active sessions
See Execution Model for details on retries, request IDs, and key revocation behavior.
Authentication Modes
TengineAI supports two client authentication modes:
| Mode | Token | Use Case |
|---|---|---|
| API Key | tengine_... | Workflows, automation, server-to-server |
| User-Scoped | tng_mst_... | Multi-tenant apps, per-user tool execution |
Both modes pass the credential as the authorization_token in the Anthropic SDK's mcp_servers configuration.
See API Keys and User-Scoped Sessions for full details.
What TengineAI Is Not
TengineAI is not:
- A prebuilt integration catalog (no Gmail connector, no Reddit connector, no SaaS OAuth flows)
- An agent framework
- A model hosting service
- A workflow orchestrator
- A prompt management system
It is execution infrastructure. You define the tools. The model decides what to call. TengineAI executes it safely.
The model can only call tools you have explicitly registered and enabled. TengineAI does not fetch arbitrary URLs — every endpoint the model can reach was configured by you.
Next Steps
- Get Started in 5 Minutes – Register a tool and run your first model-invoked API call
- Architecture Overview – Understand the full execution flow
- Execution Model – Retries, request IDs, key revocation
- API Keys – Project-scoped authentication
- User-Scoped Sessions – Per-user session tokens for multi-tenant apps
- Custom Tools – How to define and configure tools
Serveurs connexes
DICOM MCP Server
Enables AI assistants to query, read, and move data on DICOM servers like PACS and VNA.
Baozi Bet Prediction Markets
Baozi bet MCP server to allow agents create their own prediction markets
OctoEverywhere For 3D Printing
A 3D Printing MCP server that allows for querying for live state, webcam snapshots, and 3D printer control.
Decompose
Decompose text into classified semantic units — authority, risk, attention, entities. No LLM. Deterministic.
Microsoft Learn MCP Server
The Microsoft Learn MCP Server enables clients like GitHub Copilot and other AI agents to bring trusted and up-to-date information directly from Microsoft's official documentation. It is a remote MCP server that uses streamable http. It allows to search through documentation, fetch a complete article, and search through code samples.
Shioaji MCP Server
Access the Shioaji trading API for financial data and trading operations, requiring a SinoPac Securities account.
A Christmas Carol by Charles Dickens
Semantic search through Dickens' classic tale. Find passages by meaning, theme, or concept - not just keywords.
Crypto Trader
Provides real-time cryptocurrency market data using the CoinGecko API.
VMS Integration
Connects to a CCTV recording program (VMS) to retrieve recorded and live video streams and control the VMS software.
AFL (Australian Football League)
Provides Australian Football League (AFL) data, including games, standings, and team information, from the Squiggle API.