AI-NATIVE OPERATIONS

Your fleet, controlled by intent

Connect Claude, GPT-4, or any MCP client to TucDesk's fleet API. Run natural language commands with human-in-the-loop approval gates.

MCP INTEGRATION

12 native MCP tools for AI agents

•list_agents — enumerate online machines with metadata
•get_agent — fetch full agent record and security posture
•connect_session — initiate an authenticated terminal session
•run_command — execute a command with approval gate
•list_sessions — recent session history with recordings
•get_audit_log — tamper-evident action history
•list_fleet_runs — parallel execution history
•get_recording — retrieve session recording metadata
•list_address_book — access controlled endpoint registry
•get_security_posture — per-agent compliance snapshot
•manage_acl — read/write access control policies
•pair_agent — complete agent onboarding from MCP

{
  "mcpServers": {
    "tucdesk": {
      "command": "npx",
      "args": ["-y", "tucdesk-mcp"],
      "env": {
        "TUCDESK_API_URL": "https://api.tucdesk.app",
        "TUCDESK_API_KEY": "tdk_..."
      }
    }
  }
}█

TUC-C LANGUAGE

Natural language to validated execution plan

•Input parsing identifies the operator intent, target set, and requested action.
•Risk assessment classifies the plan as CRITICAL, HIGH, MEDIUM, or LOW.
•HIGH and CRITICAL plans require explicit operator approval before execution.
•LOW and MEDIUM plans execute only after ACL evaluation and consent checks pass.
•Post-execution results are written to the immutable audit log.

MCP TOOL REFERENCE

Available tools, scopes, and shapes

Tool	Input parameters	Output shape	Example use	Scope
`list_agents`	team_id, filters	agents[], online_count	Read fleet status	agents:read
`get_agent`	agent_id	agent, posture, tags	Inspect one machine	agents:read
`connect_session`	agent_id, mode	session_id, consent_state	Start a session	sessions:write
`run_command`	agent_id/tag, command	run_id, risk, approval	Execute approved command	fleet:execute
`list_sessions`	cursor, limit	sessions[], next_cursor	Review history	sessions:read
`get_audit_log`	actor, action, cursor	audit_entries[]	Compliance review	audit:read
`list_fleet_runs`	tag, status	runs[]	Review batch execution	fleet:read
`get_recording`	recording_id	metadata, signed_url	Fetch playback metadata	recordings:read
`list_address_book`	query	endpoints[]	Find curated endpoint	address_book:read
`get_security_posture`	agent_id	checks[], score	Check compliance	security:read
`manage_acl`	policy_id, rule	policy, decision	Read or update policy	acl:write
`pair_agent`	pairing_key	agent_id, status	Complete onboarding	agents:write

SELF-HOSTED MCP

Point AI clients at your own API

Self-hosted deployments use the same MCP server. The only difference is the API URL and API key stored in the MCP client configuration.

{
  "mcpServers": {
    "tucdesk": {
      "command": "npx",
      "args": ["-y", "tucdesk-mcp"],
      "env": {
        "TUCDESK_API_URL": "https://api.yourdomain.com",
        "TUCDESK_API_KEY": "tdk_self_hosted_..."
      }
    }
  }
}█

SAFETY MODEL

AI agents cannot bypass the consent gate

request pipelineidentity → ACL → risk → approval → execution → audit

Every AI action is evaluated like a human operator action. The model favors bounded tools, explicit target scope, and auditable decisions over autonomous shell access.

1. Scope identity

The MCP request is bound to a signed operator identity, team_id, tool scope, and target set.

2. Parse intent

TUC-C extracts action, target, command arguments, and confidence before anything reaches an agent.

3. Classify risk

Risk tiers decide whether the request can proceed directly or must wait for human confirmation.

4. Write evidence

Execution result, approval state, actor, target, and command context are recorded for audit review.

HIGH and CRITICAL actions require explicit human confirmation. LOW and MEDIUM actions still pass ACL, consent state, target validation, and immutable audit logging before results are returned to the AI client.