API Endpoints

Agent Runtimes exposes a comprehensive REST API for managing agents, executing prompts, and monitoring system status.

Base URL

http://localhost:8765/api/v1

URL Convention - Trailing Slashes

Most API endpoints should be called without trailing slashes. However, mounted protocol endpoints require trailing slashes:

Standard endpoints (no trailing slash):

✅ /api/v1/agents
✅ /api/v1/mcp/servers

Protocol endpoints (trailing slash required):

✅ /api/v1/ag-ui/{agent_id}/ - AG-UI protocol (mounted Starlette apps)
✅ /api/v1/a2a/{agent_id}/ - A2A protocol (FastA2A compatibility)
✅ /api/v1/examples/{example_name}/ - AG-UI example agents

Without the trailing slash, mounted apps return a 307 redirect which may break streaming clients.

Agents

Manage AI agents and their configurations.

List Agents

GET /api/v1/agents

Returns a list of all registered agents.

Response:

{
  "agents": [
    {
      "id": "pydantic-ai",
      "name": "Pydantic AI Agent",
      "transport": "streaming",
      "model": "gpt-4o"
    }
  ]
}

Get Agent

GET /api/v1/agents/{agent_id}

Get details for a specific agent.

Response:

{
  "id": "pydantic-ai",
  "name": "Pydantic AI Agent",
  "transport": "streaming",
  "model": "gpt-4o",
  "mcp_toolsets": ["tavily", "fetch"]
}

Create Agent

POST /api/v1/agents
Content-Type: application/json

Create a new agent.

Request Body:

{
  "id": "my-agent",
  "name": "My Custom Agent",
  "model": "gpt-4o",
  "system_prompt": "You are a helpful assistant."
}

Update Agent

PUT /api/v1/agents/{agent_id}
Content-Type: application/json

Update an existing agent configuration.

Update Agent MCP Servers (Runtime)

PATCH /api/v1/agents/{agent_id}/mcp-servers
Content-Type: application/json

Dynamically update an agent's selected MCP servers without recreating the agent. This allows adding or removing MCP tools from a running agent.

Request Body:

{
  "selected_mcp_servers": ["tavily", "filesystem"]
}

Response:

{
  "agent_id": "my-agent",
  "selected_mcp_servers": ["tavily", "filesystem"],
  "message": "MCP servers updated successfully"
}

Start Agent MCP Servers

Start catalog MCP servers defined for running agents. Environment variables can be provided to configure the servers (e.g., API keys). If the agent has Codemode enabled, the Codemode toolset will be rebuilt to include the newly started servers as programmatic tools.

Two-Phase Sandbox Bootstrap (Kubernetes)

In a Kubernetes pod, agent-runtimes starts before the Jupyter container is ready. The code sandbox goes through two phases:

Phase	Trigger	Sandbox Variant	Code Execution
1 — Startup	Dockerfile CMD (`--codemode`, no `--jupyter-sandbox`)	`local-eval`	In-process `eval()` (fallback)
2 — Companion call	`POST /api/v1/agents/mcp-servers/start` with `jupyter_sandbox` + `mcp_proxy_url`	`local-jupyter`	Jupyter kernel (`:2300`)

Phase 1: The agent-runtimes container starts with --codemode and --no-catalog-mcp-servers. Because no --jupyter-sandbox flag is provided, the CodeSandboxManager defaults to variant="local-eval".

Phase 2: Once all containers in the pod are healthy, the runtimes-companion sidecar calls this endpoint with:

jupyter_sandbox: http://127.0.0.1:2300?token=<token> — the Jupyter kernel URL
mcp_proxy_url: http://127.0.0.1:8765/api/v1/mcp/proxy — the HTTP-to-stdio proxy

The endpoint calls CodeSandboxManager.configure_from_url(), which switches the variant to local-jupyter and rebuilds the Codemode toolset. All subsequent code execution runs inside the Jupyter kernel instead of local eval.

┌──────────────────────────────────────────────────────────────────────┐
│  Pod                                                                 │
│                                                                      │
│  companion ──POST /mcp-servers/start──▶ agent-runtimes :8765         │
│              (jupyter_sandbox,           │                            │
│               mcp_proxy_url,             ├─ starts MCP servers       │
│               env_vars)                  ├─ switches to local-jupyter │
│                                          └─ rebuilds codemode        │
│                                                                      │
│  agent-runtimes :8765 ──run_code()──▶ jupyter :2300                  │
│  jupyter :2300 ──HTTP tool call──▶ agent-runtimes /api/v1/mcp/proxy  │
│                                                                      │
│  Shared Volume: /mnt/shared-agent/                                   │
│  ├── generated/  (Python tool bindings written by codemode)          │
│  └── skills/     (SKILL.md files for agent skills)                   │
└──────────────────────────────────────────────────────────────────────┘

Start MCP Servers for All Agents

POST /api/v1/agents/mcp-servers/start
Content-Type: application/json

Start catalog MCP servers for all running agents.

Request Body:

{
  "env_vars": [
    {"name": "TAVILY_API_KEY", "value": "tvly-xxx"},
    {"name": "GITHUB_TOKEN", "value": "ghp_xxx"}
  ],
  "jupyter_sandbox": "http://localhost:8888?token=my-token",
  "mcp_proxy_url": "http://127.0.0.1:8765/api/v1/mcp/proxy"
}

Field	Type	Required	Description
`env_vars`	array	No	List of environment variables to set before starting servers
`env_vars[].name`	string	Yes	Environment variable name
`env_vars[].value`	string	Yes	Environment variable value
`jupyter_sandbox`	string	No	Jupyter server URL with token for code execution. When provided, switches the sandbox from `local-eval` to `local-jupyter` (Phase 2 of the bootstrap)
`mcp_proxy_url`	string	No	HTTP proxy URL the Jupyter kernel uses to call MCP tools. Defaults to `http://127.0.0.1:8765/api/v1/mcp/proxy` when `jupyter_sandbox` is set

Example Request:

curl -X POST http://localhost:8000/api/v1/agents/mcp-servers/start \
  -H "Content-Type: application/json" \
  -d '{"env_vars": [{"name": "TAVILY_API_KEY", "value": "tvly-xxx"}]}'

Response: 200 OK

{
  "agents_processed": ["agent-1", "agent-2"],
  "started_servers": ["tavily", "github"],
  "stopped_servers": [],
  "already_running": [],
  "already_stopped": [],
  "failed_servers": [],
  "codemode_rebuilt": true,
  "sandbox_configured": true,
  "sandbox_variant": "local-jupyter",
  "message": "Started 2 server(s) across 2 agent(s), sandbox=local-jupyter"
}

Start MCP Servers for a Specific Agent

POST /api/v1/agents/{agent_id}/mcp-servers/start
Content-Type: application/json

Start catalog MCP servers for a specific running agent.

Path Parameters:

Parameter	Type	Description
`agent_id`	string	The agent identifier

Request Body:

{
  "env_vars": [
    {"name": "TAVILY_API_KEY", "value": "tvly-xxx"},
    {"name": "GITHUB_TOKEN", "value": "ghp_xxx"}
  ]
}

Field	Type	Required	Description
`env_vars`	array	No	List of environment variables to set before starting servers
`env_vars[].name`	string	Yes	Environment variable name
`env_vars[].value`	string	Yes	Environment variable value

Example Request:

curl -X POST http://localhost:8000/api/v1/agents/my-agent/mcp-servers/start \
  -H "Content-Type: application/json" \
  -d '{"env_vars": [{"name": "TAVILY_API_KEY", "value": "tvly-xxx"}]}'

Response: 200 OK

{
  "agent_id": "my-agent",
  "started_servers": ["tavily"],
  "stopped_servers": [],
  "already_running": ["github"],
  "already_stopped": [],
  "failed_servers": [
    {"server_id": "filesystem", "error": "Server config not found"}
  ],
  "codemode_rebuilt": true,
  "message": "Started 1 server(s), 1 already running, 1 failed"
}

Response Fields

Field	Type	Description
`agent_id`	string	The agent identifier (only for single-agent endpoint)
`agents_processed`	array	List of agent IDs that were processed (only for all-agents endpoint)
`started_servers`	array	List of server IDs that were successfully started
`already_running`	array	List of server IDs that were already running
`failed_servers`	array	List of servers that failed to start with error details
`codemode_rebuilt`	boolean	Whether the Codemode toolset was rebuilt
`sandbox_configured`	boolean	Whether the code sandbox was (re)configured
`sandbox_variant`	string	The sandbox variant after configuration (`local-eval` or `local-jupyter`)
`message`	string	Summary message

Error Responses:

Code	Description
`404`	Agent not found (single-agent endpoint only)
`500`	Failed to start MCP servers

Stop Agent MCP Servers

Stop catalog MCP servers for running agents.

Stop MCP Servers for All Agents

POST /api/v1/agents/mcp-servers/stop

Stop catalog MCP servers for all running agents.

Example Request:

curl -X POST http://localhost:8000/api/v1/agents/mcp-servers/stop

Response: 200 OK

{
  "agents_processed": ["agent-1", "agent-2"],
  "started_servers": [],
  "stopped_servers": ["tavily", "github"],
  "already_running": [],
  "already_stopped": [],
  "failed_servers": [],
  "codemode_rebuilt": false,
  "message": "Stopped 2 server(s) across 2 agent(s)"
}

Stop MCP Servers for a Specific Agent

POST /api/v1/agents/{agent_id}/mcp-servers/stop

Stop catalog MCP servers for a specific running agent.

Path Parameters:

Parameter	Type	Description
`agent_id`	string	The agent identifier

Example Request:

curl -X POST http://localhost:8000/api/v1/agents/my-agent/mcp-servers/stop

Response: 200 OK

{
  "agent_id": "my-agent",
  "started_servers": [],
  "stopped_servers": ["tavily", "github"],
  "already_running": [],
  "already_stopped": ["filesystem"],
  "failed_servers": [],
  "codemode_rebuilt": false,
  "message": "Stopped 2 server(s), 1 already stopped"
}

Response Fields

Field	Type	Description
`agent_id`	string	The agent identifier (only for single-agent endpoint)
`agents_processed`	array	List of agent IDs that were processed (only for all-agents endpoint)
`stopped_servers`	array	List of server IDs that were successfully stopped
`already_stopped`	array	List of server IDs that were already stopped
`failed_servers`	array	List of servers that failed to stop with error details
`message`	string	Summary message

Error Responses:

Code	Description
`404`	Agent not found (single-agent endpoint only)
`500`	Failed to stop MCP servers

Code Sandbox Configuration

The Code Sandbox Manager controls how code is executed when using Code Mode or Skills. It supports two variants:

local-eval: Uses Python exec() for simple code execution (default)
local-jupyter: Connects to an external Jupyter server for persistent kernel state

Get Sandbox Status

GET /api/v1/agents/sandbox/status

Get the current status of the code sandbox manager.

Example Request:

curl http://localhost:8000/api/v1/agents/sandbox/status

Response: 200 OK

{
  "variant": "local-eval",
  "jupyter_url": null,
  "jupyter_token_set": false,
  "sandbox_running": true
}

Field	Type	Description
`variant`	string	Current sandbox variant (`local-eval` or `local-jupyter`)
`jupyter_url`	string	Jupyter server URL if configured
`jupyter_token_set`	boolean	Whether a Jupyter token is configured
`sandbox_running`	boolean	Whether a sandbox instance is currently active

Configure Sandbox

POST /api/v1/agents/sandbox/configure
Content-Type: application/json

Configure the code sandbox manager at runtime. If a sandbox is running with a different configuration, it will be stopped and recreated on next use.

Request Body:

{
  "variant": "local-jupyter",
  "jupyter_url": "http://localhost:8888?token=my-token"
}

Field	Type	Required	Description
`variant`	string	No	Sandbox variant: `local-eval` (default) or `local-jupyter`
`jupyter_url`	string	Required for `local-jupyter`	Jupyter server URL (can include token as query param)
`jupyter_token`	string	No	Jupyter token (overrides token in URL if provided)

Example Requests:

# Configure for Jupyter sandbox
curl -X POST http://localhost:8000/api/v1/agents/sandbox/configure \
  -H "Content-Type: application/json" \
  -d '{"variant": "local-jupyter", "jupyter_url": "http://localhost:8888?token=my-token"}'

# Reset to local-eval sandbox
curl -X POST http://localhost:8000/api/v1/agents/sandbox/configure \
  -H "Content-Type: application/json" \
  -d '{"variant": "local-eval"}'

Response: 200 OK

{
  "variant": "local-jupyter",
  "jupyter_url": "http://localhost:8888",
  "jupyter_token_set": true,
  "sandbox_running": false
}

Error Responses:

Code	Description
`400`	`jupyter_url` is required when variant is `local-jupyter`
`500`	Failed to configure sandbox

Restart Sandbox

POST /api/v1/agents/sandbox/restart

Restart the code sandbox with current configuration. This stops any running sandbox and creates a new instance.

Example Request:

curl -X POST http://localhost:8000/api/v1/agents/sandbox/restart

Response: 200 OK

{
  "variant": "local-jupyter",
  "jupyter_url": "http://localhost:8888",
  "jupyter_token_set": true,
  "sandbox_running": true
}

Error Responses:

Code	Description
`500`	Failed to restart sandbox

Delete Agent

DELETE /api/v1/agents/{agent_id}

Remove an agent from the system.

Prompts

Execute prompts against agents and manage conversations.

Send Prompt (Streaming)

POST /api/v1/agents/{agent_id}/prompt
Content-Type: application/json

Send a prompt to an agent and receive a streaming response.

Request Body:

{
  "message": "What is the weather in Paris?",
  "conversation_id": "optional-conversation-id"
}

Response: Server-Sent Events (SSE) stream

data: {"type": "text", "content": "Based on "}
data: {"type": "text", "content": "the current weather..."}
data: {"type": "tool_call", "name": "get_weather", "args": {...}}
data: {"type": "tool_result", "result": {...}}
data: {"type": "done"}

Send Prompt (Non-Streaming)

POST /api/v1/agents/{agent_id}/prompt/sync
Content-Type: application/json

Send a prompt and wait for the complete response.

Response:

{
  "response": "Based on the current weather data...",
  "tool_calls": [...],
  "usage": {
    "prompt_tokens": 150,
    "completion_tokens": 200
  }
}

Configuration

System configuration and status endpoints.

Get Configuration

GET /api/v1/configure

Get the current system configuration including available models and MCP servers.

Response:

{
  "models": [...],
  "mcp_servers": [...],
  "tools": [...]
}

MCP Toolsets Status

GET /api/v1/configure/mcp-toolsets-status

Get the status of MCP toolsets initialization.

Response:

{
  "initialized": true,
  "ready_count": 3,
  "failed_count": 0,
  "ready_servers": ["tavily", "linkedin", "kaggle"],
  "failed_servers": {}
}

MCP Toolsets Info

GET /api/v1/configure/mcp-toolsets-info

Get detailed information about running MCP toolsets.

Response:

[
  {
    "type": "MCPServerStdio",
    "id": "tavily",
    "command": "npx",
    "args": ["-y", "tavily-mcp@0.1.3"]
  }
]

MCP Servers

Manage MCP (Model Context Protocol) servers for tool integration. There are two types of MCP servers:

MCP Config: User-defined servers from ~/.datalayer/mcp.json that start automatically
MCP Catalog: Predefined servers that can be enabled on-demand

Separate Storage

Config and catalog servers are stored separately, allowing the same server ID to exist in both without conflict. For example, you can have a custom tavily server in your mcp.json while also having access to the predefined tavily in the catalog.

Starting Without Config MCP Servers

When you start the server with the --no-config-mcp-servers CLI flag, config MCP servers from ~/.datalayer/mcp.json are not started automatically. You can then use the endpoints below to dynamically enable MCP servers from the catalog at runtime:

# Start server without config MCP servers
python -m agent_runtimes --no-config-mcp-servers

# Then enable MCP servers via API
curl -X POST http://localhost:8000/api/v1/mcp/servers/catalog/tavily/enable
curl -X POST http://localhost:8000/api/v1/mcp/servers/catalog/github/enable

This is useful for scenarios where you want fine-grained control over which MCP servers are running, or when you need to manage resources carefully.

List MCP Config Servers

GET /api/v1/mcp/servers/config

Get all running MCP Config servers from ~/.datalayer/mcp.json. These servers start automatically when the agent runtime starts.

Response:

[
  {
    "id": "tavily",
    "name": "Tavily Search",
    "description": "Web search and research capabilities",
    "enabled": true,
    "tools": [
      {
        "name": "tavily-search",
        "description": "Search the web using Tavily",
        "enabled": true
      }
    ],
    "isAvailable": true,
    "isRunning": true,
    "isConfig": true,
    "transport": "stdio"
  }
]

List Catalog Servers

GET /api/v1/mcp/servers/catalog

Get all predefined MCP servers from the catalog. These are NOT started automatically.

Response:

[
  {
    "id": "tavily",
    "name": "Tavily Search",
    "description": "Web search and research capabilities via Tavily API",
    "enabled": true,
    "tools": [],
    "command": "npx",
    "args": ["-y", "tavily-mcp"],
    "requiredEnvVars": ["TAVILY_API_KEY"],
    "isAvailable": false,
    "isConfig": false,
    "transport": "stdio"
  }
]

List All Available Servers

GET /api/v1/mcp/servers/available

Get all available MCP servers - combines catalog servers with running config servers. Since config and catalog servers are stored separately, the same ID can appear in both (as separate entries).

Response:

[
  {
    "id": "tavily",
    "name": "Tavily Search",
    "isAvailable": true,
    "isRunning": true,
    "isConfig": true,
    "transport": "stdio"
  },
  {
    "id": "tavily",
    "name": "Tavily Search",
    "description": "Web search via Tavily API",
    "isAvailable": false,
    "isRunning": false,
    "isConfig": false,
    "transport": "stdio"
  }
]

Note: The same ID (tavily) can appear twice - once from config (user's mcp.json) and once from catalog (predefined).

List Running Servers

GET /api/v1/mcp/servers

Get all currently running MCP servers (both config and catalog).

Response:

[
  {
    "id": "tavily",
    "name": "Tavily Search",
    "enabled": true,
    "tools": [
      {
        "name": "tavily-search",
        "description": "A powerful web search tool...",
        "enabled": true,
        "inputSchema": {...}
      }
    ],
    "isAvailable": true
  }
]

Get Server

GET /api/v1/mcp/servers/{server_id}

Get details for a specific MCP server.

Enable Catalog Server

POST /api/v1/mcp/servers/catalog/{server_name}/enable

Start an MCP server from the catalog for the current session. Only works for catalog servers (not config servers which start automatically).

This is particularly useful when:

You started the server with --no-config-mcp-servers
You want to dynamically add tools to an agent at runtime
You need fine-grained control over which MCP servers are running

Path Parameters:

Parameter	Type	Description
`server_name`	string	The name/ID of the MCP server from the catalog (e.g., `tavily`, `github`, `filesystem`)

Example Request:

curl -X POST http://localhost:8000/api/v1/mcp/servers/catalog/tavily/enable

Response: 201 Created

{
  "id": "tavily",
  "name": "Tavily Search",
  "enabled": true,
  "tools": [
    {
      "name": "tavily-search",
      "description": "Search the web using Tavily"
    }
  ],
  "isRunning": true,
  "isConfig": false
}

Error Responses:

Code	Description
`404`	Server not found in catalog
`500`	Failed to start the MCP server

Disable Catalog Server

DELETE /api/v1/mcp/servers/catalog/{server_name}/disable

Stop an MCP server and remove it from the current session. This stops the MCP server process and frees up resources.

Path Parameters:

Parameter	Type	Description
`server_name`	string	The name/ID of the MCP server to disable

Example Request:

curl -X DELETE http://localhost:8000/api/v1/mcp/servers/catalog/tavily/disable

Response: 204 No Content

Error Responses:

Code	Description
`404`	Server is not currently enabled
`500`	Failed to stop the MCP server

Create Server

POST /api/v1/mcp/servers
Content-Type: application/json

Add a new MCP server configuration.

Request Body:

{
  "id": "my-server",
  "name": "My Custom Server",
  "command": "npx",
  "args": ["-y", "my-mcp-server"],
  "enabled": true
}

Update Server

PUT /api/v1/mcp/servers/{server_id}
Content-Type: application/json

Update an existing MCP server configuration.

Delete Server

DELETE /api/v1/mcp/servers/{server_id}

Remove an MCP server configuration.

Response: 204 No Content

AG-UI Protocol

The AG-UI (Agent-User Interface) protocol provides streaming agent communication using Server-Sent Events (SSE). Each agent is mounted as a Starlette application.

Trailing Slash Required

AG-UI endpoints are mounted Starlette apps and require a trailing slash. Without it, you'll get a 307 redirect that breaks streaming.

List AG-UI Agents

GET /api/v1/ag-ui/agents

List all agents available via AG-UI protocol.

Response:

{
  "protocol": "ag-ui",
  "version": "0.1.0",
  "agents": [
    {
      "agent_id": "pydantic-ai",
      "endpoint": "/api/v1/ag-ui/pydantic-ai/"
    }
  ],
  "agents_endpoint": "/api/v1/ag-ui/agents",
  "terminate_endpoint": "/api/v1/ag-ui/terminate",
  "note": "Each agent is mounted at /api/v1/ag-ui/{agent_id}/ (trailing slash required)"
}

AG-UI Protocol Info

GET /api/v1/ag-ui/

Get AG-UI protocol information and list of available agents.

Send Message (Streaming)

POST /api/v1/ag-ui/{agent_id}/
Content-Type: application/json

Send a message to an AG-UI agent and receive streaming SSE response.

Request Body:

{
  "thread_id": "thread-123",
  "run_id": "run-456",
  "messages": [
    {
      "id": "msg-1",
      "role": "user",
      "content": "Hello, how are you?"
    }
  ],
  "state": {},
  "tools": [],
  "context": [],
  "forwardedProps": {}
}

Response: Server-Sent Events (SSE) stream with AG-UI events:

event: RUN_STARTED
data: {"type": "RUN_STARTED", "thread_id": "thread-123", "run_id": "run-456"}

event: TEXT_MESSAGE_START
data: {"type": "TEXT_MESSAGE_START", "message_id": "msg-2", "role": "assistant"}

event: TEXT_MESSAGE_CONTENT
data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg-2", "delta": "Hello! "}

event: TEXT_MESSAGE_CONTENT
data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg-2", "delta": "I'm doing well."}

event: TEXT_MESSAGE_END
data: {"type": "TEXT_MESSAGE_END", "message_id": "msg-2"}

event: RUN_FINISHED
data: {"type": "RUN_FINISHED", "thread_id": "thread-123", "run_id": "run-456"}

Terminate Run

POST /api/v1/ag-ui/terminate
Content-Type: application/json

Terminate an active AG-UI run.

Request Body:

{
  "thread_id": "thread-123",
  "run_id": "run-456"
}

AG-UI Event Types

Event	Description
`RUN_STARTED`	Run has started processing
`RUN_FINISHED`	Run completed successfully
`RUN_ERROR`	Run encountered an error
`TEXT_MESSAGE_START`	Beginning of a text message
`TEXT_MESSAGE_CONTENT`	Incremental text content (streaming)
`TEXT_MESSAGE_END`	End of a text message
`TOOL_CALL_START`	Tool invocation started
`TOOL_CALL_ARGS`	Tool call arguments (streaming)
`TOOL_CALL_END`	Tool invocation completed
`TOOL_CALL_RESULT`	Result from tool execution
`STATE_SNAPSHOT`	Current state snapshot
`STATE_DELTA`	Incremental state update
`CUSTOM`	Custom event type

Example: Using curl

curl -X POST "http://localhost:8765/api/v1/ag-ui/pydantic-ai/" \
  -H "Content-Type: application/json" \
  -d '{
    "thread_id": "test-thread",
    "run_id": "test-run",
    "messages": [{"id": "1", "role": "user", "content": "Hello!"}],
    "state": {},
    "tools": [],
    "context": [],
    "forwardedProps": {}
  }'

Example: Using TypeScript

import { AGUIAdapter } from '@datalayer/agent-runtimes';

const adapter = new AGUIAdapter({
  baseUrl: 'http://localhost:8765/api/v1/ag-ui/pydantic-ai/',
  // Note: The adapter automatically ensures trailing slash
});

await adapter.sendMessage('Hello!', {
  onToken: (token) => console.log(token),
  onComplete: (message) => console.log('Done:', message),
});

Example: Using Python

from agent_runtimes.transports.clients import AGUIClient

async with AGUIClient("http://localhost:8765/api/v1/ag-ui/pydantic-ai/") as client:
    async for event in client.run("Hello!"):
        print(event)

Extensions

A2UI Endpoints

GET  /api/v1/a2ui/           # A2UI protocol
GET  /api/v1/a2ui/agents     # List A2UI agents

MCP-UI Endpoints

GET  /api/v1/mcp-ui/         # MCP-UI protocol
GET  /api/v1/mcp-ui/agents   # List MCP-UI agents

Conversations

Manage conversation history and context.

Get Conversation

GET /api/v1/conversations/{conversation_id}

Retrieve a conversation by ID.

List Conversations

GET /api/v1/conversations

List all conversations, optionally filtered by agent.

Delete Conversation

DELETE /api/v1/conversations/{conversation_id}

Delete a conversation and its history.

Health & Status

Health Check

GET /api/v1/health

Response:

{
  "status": "healthy",
  "version": "0.4.5"
}

Readiness Check

GET /api/v1/ready

Returns 200 when the service is ready to accept requests.

Error Responses

All endpoints return standard error responses:

{
  "detail": "Agent not found",
  "status_code": 404
}

Common Status Codes

Code	Description
`200`	Success
`201`	Created
`400`	Bad Request
`404`	Not Found
`422`	Validation Error
`500`	Internal Server Error

Authentication

info

Authentication is optional and configurable. When enabled, include the API key in the request header:

Authorization: Bearer your-api-key

Rate Limiting

Rate limiting can be configured per endpoint. When rate limited, you'll receive:

HTTP/1.1 429 Too Many Requests
Retry-After: 60

OpenAPI Documentation

Interactive API documentation is available at:

Swagger UI: http://localhost:8765/docs
ReDoc: http://localhost:8765/redoc
OpenAPI JSON: http://localhost:8765/openapi.json

Base URL​

Agents​

List Agents​

Get Agent​

Create Agent​

Update Agent​

Update Agent MCP Servers (Runtime)​

Start Agent MCP Servers​

Start MCP Servers for All Agents​

Start MCP Servers for a Specific Agent​

Response Fields​

Stop Agent MCP Servers​

Stop MCP Servers for All Agents​

Stop MCP Servers for a Specific Agent​

Response Fields​

Code Sandbox Configuration​

Get Sandbox Status​

Configure Sandbox​

Restart Sandbox​

Delete Agent​

Prompts​

Send Prompt (Streaming)​

Send Prompt (Non-Streaming)​

Configuration​

Get Configuration​

MCP Toolsets Status​

MCP Toolsets Info​

MCP Servers​

List MCP Config Servers​

List Catalog Servers​

List All Available Servers​

List Running Servers​

Get Server​

Enable Catalog Server​

Disable Catalog Server​

Create Server​

Update Server​

Delete Server​

AG-UI Protocol​

List AG-UI Agents​

AG-UI Protocol Info​

Send Message (Streaming)​

Terminate Run​

AG-UI Event Types​

Example: Using curl​

Example: Using TypeScript​

Example: Using Python​

Extensions​

A2UI Endpoints​

MCP-UI Endpoints​

Conversations​

Get Conversation​

List Conversations​

Delete Conversation​

Health & Status​

Health Check​

Readiness Check​

Error Responses​

Common Status Codes​

Authentication​

Rate Limiting​

OpenAPI Documentation​

Base URL

Agents

List Agents

Get Agent

Create Agent

Update Agent

Update Agent MCP Servers (Runtime)

Start Agent MCP Servers

Start MCP Servers for All Agents

Start MCP Servers for a Specific Agent

Response Fields

Stop Agent MCP Servers

Stop MCP Servers for All Agents

Stop MCP Servers for a Specific Agent

Response Fields

Code Sandbox Configuration

Get Sandbox Status

Configure Sandbox

Restart Sandbox

Delete Agent

Prompts

Send Prompt (Streaming)

Send Prompt (Non-Streaming)

Configuration

Get Configuration

MCP Toolsets Status

MCP Toolsets Info

MCP Servers

List MCP Config Servers

List Catalog Servers

List All Available Servers

List Running Servers

Get Server

Enable Catalog Server

Disable Catalog Server

Create Server

Update Server

Delete Server

AG-UI Protocol

List AG-UI Agents

AG-UI Protocol Info

Send Message (Streaming)

Terminate Run

AG-UI Event Types

Example: Using curl

Example: Using TypeScript

Example: Using Python

Extensions

A2UI Endpoints

MCP-UI Endpoints

Conversations

Get Conversation

List Conversations

Delete Conversation

Health & Status

Health Check

Readiness Check

Error Responses

Common Status Codes

Authentication

Rate Limiting

OpenAPI Documentation