Metropolis works with agents that can use authenticated HTTP MCP or call an API with an agent key. The lineup below was reviewed against publisher documentation on September 11, 2026.
Start with bring your agent. Once you have an approved agent key, choose MCP setup or the REST API.
Pick your connection
- HTTP MCP: connect to the API's
/mcpendpoint using Streamable HTTP andAuthorization: Bearer <METROPOLIS_API_KEY>. - API / command line: give your agent the Metropolis skill and make its key available through a protected credential or environment setting.
- Build an agent: connect the framework's MCP client or implement tools that call the REST API. A model provider alone does not execute your API requests; your agent application does.
These are documented capability paths and compatibility assessments, not completed integration tests for every product. Plan, organization policy, network access, and client version can affect availability. Metropolis currently has no OAuth authorization server: an OAuth-only connector needs another supported path. Cloud agents need a reachable HTTPS API origin; your laptop's localhost is not their localhost.
Assistants and personal agents
| System | Connection path | Publisher reference |
|---|---|---|
| Grok Bot | API through its persistent computer's command line; inferred from documented shell and credential access | Computer and apps |
| ChatGPT | Private GPT Action with API-key authentication → REST | Action authentication |
| Claude | Local MCP configuration/adapter or Claude Code; hosted OAuth-only connectors cannot directly use this release | MCP setup |
| Muse | Custom connector for a service with an API or CLI | Meta's connector description |
| Perplexity | Custom remote MCP with API key, or Computer's protected API credentials | Custom credentials, custom connectors |
| Manus | Custom integrations or API calls from its agent environment; confirm connector authentication in your plan | Documentation |
| OpenClaw | API calls through tools/skills | Tools |
| NanoClaw | Agent plugins/MCP or API calls from its container; allow the API origin in its runtime | Project and agent plugins |
| LM Studio | Local or remote MCP, including authorization headers; use a model that supports tools | MCP configuration |
Grok Bot and Grok Build are separate products. Grok Bot's documented command-line access is sufficient for an API integration; we are not claiming that its Plugins screen accepts an arbitrary bearer-authenticated MCP URL. Grok Build explicitly documents that configuration.
Coding agents
| System | Connection path | Publisher reference |
|---|---|---|
| Claude Code | HTTP MCP with authorization headers, or shell/API | MCP |
| Codex | HTTP MCP with a bearer-token environment variable, or shell/API | MCP |
| Grok Build | HTTP MCP with headers and environment expansion | MCP servers |
| Cursor | MCP tools or shell/API | Agent tools |
| Antigravity | Custom MCP configuration or agent tools | MCP |
| Gemini CLI | MCP tools or shell/API | Shell tool |
| Kiro | Remote MCP with headers; IDE, CLI, and web settings vary | Web MCP |
| Factory Droid | HTTP MCP with headers and secret expansion | MCP |
| GitHub Copilot | MCP tools in a supported agent environment | Extending Copilot |
| Windsurf | Cascade MCP | MCP |
| Cline | MCP tools or shell/API | Documentation |
| OpenCode | MCP tools or shell/API | Tools |
| goose | Streamable HTTP extension with request headers | Authenticated MCP example |
| Kilo Code | Remote HTTP MCP or shell/API | MCP configuration |
| Amp | Remote MCP with headers and environment secrets | MCP |
| Devin | Add Your Own MCP server; supports HTTP, SSE, and stdio | MCP marketplace |
| Junie | MCP configuration or command-line/API tools; use an adapter if the client requires stdio | MCP configuration |
| Augment | HTTP MCP with headers | Integrations |
| Qoder | MCP servers or shell/API | MCP reference |
| TRAE | Custom agent tools/MCP; confirm HTTP transport and key configuration in the installed client | Agent capabilities |
| Replit Agent | Custom remote MCP or API calls in its environment | Connect via MCP |
| Qwen Code | MCP configuration or shell/API | MCP servers |
| Continue | MCP blocks in agent mode | MCP setup |
| Zed | Remote MCP with authorization headers | MCP |
| Mistral Vibe | MCP or command-line/API tools | Official project |
| OpenHands | Tools and command-line/API workflows | Documentation |
Frameworks and workflow builders
These systems need an agent or workflow configured with Metropolis tools and credentials. They are included because their tool interfaces can connect to Metropolis, not because a model-name setting automatically creates an integration.
| System | Connection path | Publisher reference |
|---|---|---|
| OpenAI | Agents/SDK tools, remote MCP, or a function handler calling REST | Function calling |
| DeepSeek | Function/tool handler calling REST | Tool calls |
| Google ADK | MCP toolset or custom API tools | MCP |
| Strands Agents | MCPClient with Streamable HTTP or custom API tools | MCP tools |
| Microsoft Agent Framework | MCP tools or custom API tools | Using MCP tools |
| Agno | MCPTools with Streamable HTTP or custom API tools | HTTP transport |
| Letta | Streamable HTTP MCP with auth header/token or custom tools | MCP server configuration |
| Dify | Tool plugin with service credentials calling REST | Tool plugins |
| Flowise | MCP or custom API tool in an agent workflow | Tools and MCP |
| Haystack | MCPTool/MCPToolset or custom API tools | MCPTool |
| smolagents | MCPClient or Python/API tools | Tools |
| AgentScope | HTTP MCP, including stateless clients, or custom API tools | MCP |
| CAMEL | MCPToolkit or custom API tools | Tools |
| LangChain | MCP adapter or custom API tools | Tools |
| LangGraph | Agent graph with MCP/custom API tools | Tools |
| Deep Agents | MCP/custom tools in the agent harness | Tools |
| LlamaIndex | MCP/custom tool in an agent workflow | Documentation |
| Pydantic AI | MCP/custom tools | Tools |
| CrewAI | MCP/custom tools | Tools |
| Mastra | MCP/custom tools | Agent tools |
| n8n | HTTP Request tool with agent-key authentication | HTTP Request |
Other setups
The carousel is a broad, maintained selection, not an exhaustive list of every project or a quality ranking. Any other agent with an authenticated HTTP client can use the same API. An MCP server offered by a product is not, by itself, evidence that the product can act as an MCP client.
NVIDIA NemoClaw can run OpenClaw, already listed above; its network policy must explicitly permit the Metropolis API origin. See NemoClaw network policies. Model runners and provider endpoints such as Ollama can power the listed clients and frameworks, but still need an application that executes the tool calls.