MCP Server

#What is the MCP Server?

daita mcp-server starts a Model Context Protocol server over stdio. It exposes tools covering the full Daita platform — running agents, inspecting traces, managing deployments, running eval suites, reading memory, and more — so any MCP-compatible coding assistant can interact with your deployed agents directly from the editor.

#Starting the server

bash

daita mcp-server

The server runs over stdio and stays alive until the client disconnects. DAITA_API_KEY must be set in the environment or the cloud tools will return 401 errors.

#Client configuration

#Claude Desktop / Claude Code

Add the following to your claude_desktop_config.json (or .mcp.json for project-scoped config):

json

{
  "mcpServers": {
    "daita": {
      "command": "daita",
      "args": ["mcp-server"],
      "env": {
        "DAITA_API_KEY": "your_api_key"
      }
    }
  }
}

#Cursor

Add to your Cursor MCP settings:

json

{
  "mcpServers": {
    "daita": {
      "command": "daita",
      "args": ["mcp-server"],
      "env": {
        "DAITA_API_KEY": "your_api_key"
      }
    }
  }
}

Once connected, all tools listed below are available to the assistant.

#Available tools

#Agents

Tool	Required inputs	Description
`list_agents`	—	List agents. Filter by `agent_type`, `status_filter`, `page`, `per_page`
`get_agent`	`agent_id`	Get details for a specific agent
`list_deployed_agents`	—	List agents from the most recent deployments

#Deployments

Tool	Required inputs	Description
`list_deployments`	—	List active deployments for the current API key
`get_deployment_history`	`project`	Get deployment history for a project
`delete_deployment`	`deployment_id`	Delete a deployment

#Executions

Tool	Required inputs	Description
`run_agent`	`target_name`	Execute an agent or workflow and poll until complete. Accepts `target_type`, `data`, `task`, `timeout_seconds`
`list_executions`	—	List recent executions. Filter by `status`, `target_type`
`get_execution`	`execution_id`	Get details and result for an execution
`cancel_execution`	`execution_id`	Cancel a running execution
`get_execution_stats`	—	Get aggregate execution statistics

#Evals

Tool	Required inputs	Description
`list_eval_suites`	—	List registered cloud eval suites. Filter by `project_name`, `agent_name`, `status`, `page`, `per_page`
`get_eval_suite`	`eval_suite_id`	Get one registered cloud eval suite
`list_eval_runs`	—	List cloud eval run history. Filter by `eval_suite_id`, `project_name`, `status`, `page`, `per_page`
`get_eval_run`	`eval_run_id`	Get one cloud eval run summary
`get_eval_report`	`eval_run_id`	Fetch the canonical report for a cloud eval run. Accepts `detail=failures
`run_eval_suite`	one of `eval_suite_id`, `suite_name`, or `config_path`	Run a registered cloud eval suite and poll until complete. Accepts `project_name`, `timeout_seconds`, and `detail=failures

Eval report tools default to detail=failures so coding agents receive the highest-signal failure context without flooding the model with every passing case. Use detail=full when you need the complete canonical report.

#Traces

Tool	Required inputs	Description
`list_traces`	—	List traces. Filter by `status`, `agent_id`
`get_trace`	`trace_id`	Get trace details
`get_trace_spans`	`trace_id`	Get span summaries for a trace. Accepts `limit` and `include_attributes`
`get_trace_decisions`	`trace_id`	Get AI decision events recorded within a trace
`get_trace_stats`	—	Aggregate trace stats for a time period (`24h`, `7d`, `30d`)

#Schedules

Tool	Required inputs	Description
`list_schedules`	—	List agent schedules
`get_schedule`	`schedule_id`	Get schedule details
`pause_schedule`	`schedule_id`	Pause a schedule
`resume_schedule`	`schedule_id`	Resume a paused schedule

#Memory

Tool	Required inputs	Description
`get_memory_status`	`project`	Get memory system status for a project
`get_workspace_memory`	`workspace`, `project`	Get memory contents for a workspace. Defaults to a compact limit

#Secrets

Tool	Required inputs	Description
`list_secrets`	—	List stored secret key names (values are never returned)
`set_secret`	`key`, `value`	Store or update an encrypted secret
`delete_secret`	`key`	Delete a secret

#Webhooks & Conversations

Tool	Required inputs	Description
`list_webhooks`	—	List webhook URLs for the organisation

#Local development

These tools scaffold and run code locally. They require daita-agents to be installed in the same environment as the MCP server process.

Tool	Required inputs	Description
`init_project`	—	Scaffold a new Daita project. Accepts `project_name`, `project_type`
`create_agent`	`name`	Add a new agent to the current project from template
`create_workflow`	`name`	Add a new workflow to the current project from template
`create_skill`	`name`	Add a reusable skill module to the current project
`test_agent`	—	Run an agent or workflow locally and return results

#Authentication errors

If a tool returns {"error": true, "status_code": 401}, the server could not authenticate with the Daita API. Check that DAITA_API_KEY is set in the environment passed to the server process — environment variables set in your shell session are not automatically inherited by MCP servers launched by desktop apps.