[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Connect to the server WebSocket to start receiving traces. Call this first before using other Arch debug tools. Auth is automatic — stored credentials or device auth are tried in order. If device auth is needed, the browser opens automatically and the tool polls until approved (single call, no two-phase). Credentials are saved to the MCP-owned credential store for future sessions. If already connected and a new authToken is provided, the authenticated WebSocket is safely replaced before the token is committed. Changing environments while connected is rejected unless force=true is explicitly provided. Every successful response includes activeTarget identity and environment metadata; surface it to the user when context changes. If no serverUrl is given and AGENTS_URL is unset, ask the user which environment to connect to (production/dev/staging/qa) instead of guessing. If it fails, report the error as-is to the user. Do NOT try alternative approaches like REST calls.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. List all available agents from the server. Returns agents grouped by domain.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Load an agent and create a debug session. Use the agentPath format "domain/name" (e.g., "hotel-booking/booking_agent").

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Send a message to the loaded agent and optionally wait for the response.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get and search trace events. Filter by type, agent, text, error, or session. With no search filters (text/agentName/hasError), returns recent events. With search filters, searches across stored events.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. List durable historical Runtime sessions or read a bounded page of persisted trace events. Use this explicitly when live debug buffers are unavailable or historical analysis is requested. Runtime is authoritative; this tool performs one paginated request and does not retry, hydrate live stores, or silently fall back.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get the current agent state including context, gather progress, flow state, and more.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get hierarchical span tree showing execution flow. Useful for understanding agent behavior.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get all errors and warnings from the session. Includes escalations and constraint failures.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get detailed explanation of a decision event with surrounding context. Helps understand why the agent made a choice.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Get the execution graph for any agent type (scripted, reasoning, or supervisor). Shows flow steps, tools, handoffs, and routing logic. Returns JSON or Mermaid diagram format.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. List all active sessions from the server that can be subscribed to. Use this to find sessions created by the UI.

[Arch Debug] Arch connects to live sessions, traces failures, and inspects execution state. Subscribe to or unsubscribe from an existing session's trace events. Use action='subscribe' to start receiving traces (buffered + live), or action='unsubscribe' to stop. Use debug_list_active_sessions to find session IDs.

[Arch Analyze] Arch explains documentation, diagnostics, and system health signals. Get or search Agent ABL documentation from the platform. Requires platform_connect first. Provide 'topic' for full content, 'query' to search, or neither to list all available topics.

[Arch Analyze] Arch explains documentation, diagnostics, and system health signals. Get automated analysis and diagnostics for a session. Returns:

• Summary statistics (event counts, duration, LLM calls)

• Current state (step, collected fields, missing fields)

• Detected issues (loops, errors, constraint violations, tool failures)

• Suggestions for fixing problems

Use this as a starting point for debugging - it identifies common issues automatically.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Download and parse Harness CI execution logs. Returns parsed, readable log lines.

Use this to get full build/test failure logs beyond what an eval-run status summary includes.

Common usage:

• Get test failure details: stage_id="build_test", step_id="unit_tests" or "integration_tests"

• Get Docker build errors: stage_id="docker_search_ai", step_id="build_image"

• Get security scan failures: stage_id="docker_codetool_sandbox", step_id="trivy_scan"

Use the filter parameter to search for specific errors (e.g., "ECONNREFUSED|mongo|redis"). Requires HARNESS_API_KEY environment variable.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Describe the domain-neutral project-building contract, inspect a provider, inspect authoritative live project dependencies/readiness, or plan work. Workflow is the first provider; future features use the same registry and envelope.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. List, read, inspect, resume, cancel, grant, and execute durable project-building operations through a registered domain provider. Uses authoritative Studio state and attempt-bound governed actions.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage projects on the platform. Actions: list (all projects), get (by projectId), create (with name/description), update (modify name/description/entryAgentName by projectId), delete (by projectId).

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage node-based workflows end-to-end (Studio API → runtime). Actions: list, get, create, update, publish, execute, create_tool (expose a workflow as a ProjectTool for agent use), delete. Create supports every node type: start, end, function, condition, loop, delay, integration, human, data_entry, agent, tool, api. Edge sourceHandle rules: "on_success" for start/function/agent/tool/integration/api/delay/data_entry (+"on_failure" when config.onFailureEnabled); "on_approve"/"on_reject" for human; each condition id + "else" for condition; "on_complete"/"on_failure" for loop.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Drive Studio Arch SOP-build sessions through the Studio API. Actions: create_session, get_session, upload_file, send_message, continue, create_project, recover, cancel. Use create_session for SOP/onboarding or in-project Arch sessions, upload_file to upload SOP/source files, send_message to submit SOP-build instructions or fileRefs, and create_project when Arch is ready to materialize the generated project.

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Drive Arch Auto Loop repair workflows through the Studio API. Actions: list, create, get, execute_action, record_decision for project-scoped Auto Loop runs.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage project auth profiles without placing raw secrets in MCP context. Actions: list, get, create metadata, update metadata, validate, revoke, delete, providers, integrations, oauth_initiate. Complete secret entry or OAuth consent through the secure Studio flow returned by the platform.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage connector/integration connections that bind a connector to an auth profile. Actions: list, get, create, update, test, delete. Use platform_auth_profiles first when the connection requires authentication.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Provision MCP servers and turn discovered server tools into project tools. Actions: list, get, create, update, delete, test_connection, authorize, grant_status, disconnect, discover_preview, discover_import, list_tools, test_tool. Authentication is referenced by authProfileId; raw credentials are never accepted.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage agents within a project. Actions: list (all agents in project), get (agent details including DSL), save_dsl (update agent DSL). Compilation happens implicitly during version creation.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage immutable agent versions. Actions: list, get, publish (current draft with raw-DSL hash guard), qualifications, audit, and diff. Legacy create/version-promote return migration guidance without HTTP calls.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage deployments with typed version manifests. Actions: list, get, create, promote, rollback, restore, and retire. Rollback/restore/retire require explicit confirmation; qualification bypass requires a reason and existing server permission.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage the public SDK bootstrap surface. Actions: list_keys, create_key (raw public key returned once), list_channels, create_channel (binds a web/mobile/API channel to a key and deployment environment).

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage tools within a project. Actions: list (paginated tools; use page and limit to retrieve later pages), get (tool detail), create (new tool), update (modify tool), delete (remove tool), test (execute tool test with input and optional timeoutMs). Note: tool CRUD routes through the Studio API.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage Agent Tables within a project. Actions: availability, list, describe, create, update, migrate, delete, insert, query, get_row, update_row, delete_row, upsert, reveal. Use availability first to diagnose TABLE_UNAVAILABLE. Delete actions require confirm=true.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Import and export projects. Actions: export_preview (metadata preview), export (full project export as file map + manifest), import_preview (dry-run import showing changes), import (apply import). Import actions accept data.files, files, or a local folder/.zip path.

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Validate a local project folder/.zip or file map using platform-owned compiler and design diagnostics. Use in ABL repair/eval loops; returns normalized issues with suggested fixes.

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Show what the platform compiler sees in a local project package: agents, tools, handoffs, memory variables, behavior profile references, flow steps, and unresolved refs.

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Run ABL design and repair lint checks for empty RESPOND values, empty finalize steps, undeclared handoff-condition variables, side-effect tool chains, and tool+text reasoning risks.

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Given a transcript JSON and exported package folder/.zip or file map, correlate transcript failure symptoms with ABL file/line diagnoses such as finalize -> COMPLETE -> RESPOND: "".

[Arch Optimize] Arch validates packages, inspects compiler-visible models, and drives repair loops. Alias for debug_why_transcript_failed. Given transcript JSON plus project files, returns correlated ABL file/line diagnoses.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Manage eval personas through the mounted Studio API paths under /api/projects/:projectId/evals/personas. Actions: list, get, create, update, delete, templates, generate.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Manage eval scenarios through /api/projects/:projectId/evals/scenarios. Actions: list, get, create, update, delete, generate.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Manage eval evaluators through /api/projects/:projectId/evals/evaluators. Actions: list, get, create, update, delete, templates.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Manage eval sets through /api/projects/:projectId/evals/sets. Actions: list, get, create, update, delete.

[Arch Evaluate] Arch generates eval assets, runs eval workflows, and reads CI evidence. Manage eval runs for ABL repair loops. Actions: list, get, create, update, start, cancel, status, heatmap, cases, compare, preflight, quick.

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage project configuration. Actions: get_settings (project settings), update_settings (modify settings), get_llm_config (LLM configuration), update_llm_config (modify LLM config).

[Arch Build] Arch creates and changes projects, agents, tools, configuration, versions, and deployments. Manage workspaces (tenants). Actions: list (all workspaces the user belongs to, with active flag), switch (atomically authorize and switch the authenticated socket, credentials, and subsequent calls), current (show active workspace decoded from JWT). Responses include activeTarget and contextVersion; surface the active environment/workspace to the user after a switch.

[Arch Analyze] Arch explains documentation, diagnostics, and system health signals. Run diagnostic analysis on an agent or session. Returns config, findings, and suggestions. Provide sessionId for session diagnostics or agentName for agent config diagnostics. Set configOnly=true to inspect only config (model chain, credentials, tools) without running full diagnostics.