MCP tools

The hub's MCP server (apps/hub-mcp) exposes 32 tools, all named cortex_*. An agent reaches them through the cortex-hub entry the installer writes. Each call carries the API key; what a key can see depends on its project and organization scope.

This page lists each tool and its parameters. The guides explain when to use them.

Session

Tool	Parameters	What it does
cortex_session_start	repo, mode, agentId, hostname, os, ide, branch, capabilities, role, clientSessionId	Opens a session and returns the project, a mission brief, relevant knowledge and recent sessions. A clientSessionId that matches an active session reuses it
cortex_session_end	sessionId, summary, auto	Closes the session; the summary is saved as memory. auto: true is what the exit hook sends: the session is recorded as abandoned
cortex_changes	agentId, projectId	Commits and files other agents pushed that this agent has not seen
cortex_health	none	Health of the hub's backend services

See Sessions.

Code intelligence

Tool	Parameters	What it does
cortex_code_search	query, repo, projectId, org, branch, limit	Ranked hybrid search (vector and BM25) over the index. Without repo, every repo in your organization
cortex_code_context	name, repo, projectId, org, file	A symbol's callers, callees, imports and the flows it is part of
cortex_code_impact	target, repo, projectId, org, branch, direction	What a change to a symbol reaches. direction is downstream (default) or upstream
cortex_cypher	query, repo, projectId, org	A Cypher query against the code graph
cortex_detect_changes	diff, repo, projectId, org, scope, baseRef	The symbols and flows a diff touches, with a risk level. unknown means a lookup failed
cortex_list_repos	org	The indexed repos of your organization and their project ids
cortex_code_reindex	repo, branch	Starts a re-index of a branch. Run it after a push

See Code intelligence.

Knowledge and memory

Tool	Parameters	What it does
cortex_knowledge_store	title, content, tags, projectId, agentId, hallType, validFrom, files, category	Stores a document for the team. With files, it becomes a file note for those paths
cortex_knowledge_search	query, tags, projectId, limit, hallType, asOf	Semantic search over the knowledge base
cortex_file_notes	repo, projectId, files	Risks, lessons and fixes linked to files, as JSON. Without files, every linked path of the project
cortex_memory_store	content, agentId, projectId, branch, metadata	Stores a memory for this agent. branch needs projectId
cortex_memory_search	query, agentId, projectId, branch, limit	Searches memories: the branch first, then the project, then the agent
cortex_memory_delete	id	Deletes a memory that is wrong or out of date

See Knowledge and memory.

Quality

Tool	Parameters	What it does
cortex_plan_quality	plan, request, iteration, threshold, plan_type	Scores a plan on eight criteria, 0 to 10; it passes at 8.0
cortex_quality_report	gate_name, passed, score, details	Records the result of a quality gate for the dashboard
cortex_tool_stats	days, agentId, projectId	Tool usage: success rates, latency, token estimates, trends

See Quality gates.

Tasks

Tool	Parameters	What it does
cortex_task_create	title, description, assignTo, requiredCapabilities, dependsOn, notifyOnComplete, context, parentTaskId, model, effort, fallback, files, branch, reviewRequired, projectId	Creates one task, optionally for another agent
cortex_task_pickup	agentId, projectId	Tasks assigned to this agent or its API key
cortex_task_accept	taskId	Accepts an assigned task
cortex_task_update	taskId, status, message, result, parentTaskId	Moves a task to in_progress, review, completed or failed
cortex_task_list	projectId, status, assignedTo, limit	Lists tasks
cortex_task_status	taskId	A task with its logs and history
cortex_task_submit_strategy	taskId, summary, roles, subtasks, estimatedEffort	Proposes roles and subtasks for the dashboard to show for approval

Conductor

Tools for the lead agent that plans and reviews. See Conductor.

Tool	Parameters	What it does
cortex_task_tree_create	parentId, projectId, tasks, upload	Creates a whole plan as a tree of tasks; tasks refer to each other by key. upload: true returns a one-time curl command for a JSON file instead, valid for 15 minutes
cortex_agents	none	Connected workers (engines, models, busy or idle, exhausted quota) and live MCP sessions
cortex_task_review	taskId, verdict, feedback	Approves or rejects a task in review, or decides a lead gate. User gates are refused (403)
cortex_task_route	taskId, executor, model, effort, fallback, pinnedAgent, priority, files, dependsOn, clearEscalation	Changes how a task is routed. Only the fields you pass change; null clears one. dependsOn takes stored task ids
cortex_task_cancel	taskId, reason	Cancels a task; its dependents stay blocked. Cancelling a group cancels everything under it

The task fields of a plan are listed under Write the plan.

Notices in tool results

Two kinds of text can be added to the result of any hub tool call:

• Conductor notices for the lead, under 📋 Conductor:. See Review.
• File notes for the files a code tool's result points at. See File notes.