MCP (Model Context Protocol)

Tulsk uses the Model Context Protocol (MCP) as its core integration layer, enabling a standardized way to connect AI tools with your workspace data and agent infrastructure.

What is MCP?

MCP provides a standardized protocol for AI models to discover and interact with tools. Instead of custom API integrations, AI clients connect to an MCP server that advertises its capabilities, parameters, and descriptions dynamically.

Tulsk’s MCP Server

Tulsk exposes an MCP server at /api/mcp/mcp using the streamable HTTP transport. The tabs below list the tools Workspace Chat uses; an outside agent is offered the subset its token allows (see below).

Read Tools

Workspace Visibility

get_projects: List all active projects.
get_project_by_id: Fetch detailed project metadata and recent tasks.
get_task_context: Retrieve full task threads and comments.
web_search: Perform real-time web searches.
open_page: Navigate to specific Tulsk dashboard views.

Agent Awareness

Agent Insights

list_agents: Summary of all agents, skills, and health.
get_agent_detail: Deep dive into agent personas and budget.
get_agent_runs: History of agent activities filtered by status.
get_run_result: Status and output of specific autonomous runs.

Control

Execution Control

trigger_agent_run: Queue an agent for a specific task.
assign_task_to_agent: Combined assign and trigger workflow.
cancel_agent_run: Halt pending or active agent processes.

Workspace Chat PM

Project Management (Workspace Chat)

get_agent_activity_summary: Aggregated performance analytics.
get_delegation_tree: Visualize agent-to-agent collaboration.
create_workspace_schedule: Set up automated project reporting.

Actions

Data Modification

create_project / create_task: Direct workspace organization.
update_task: Modify status, priority, or assignees.
add_comment: Post updates to task threads.

Connect an outside agent

Create a token in Settings → Outside agents. A token acts as you, in one workspace, and is either Read (memory, decisions, tasks, projects, runs) or Read & write (adds writing memory, creating and updating tasks and comments, and starting agent runs). It is shown once; revoke it there at any time.

Claude Code


claude mcp add --transport http tulsk https://tulsk.io/api/mcp/mcp --header "Authorization: Bearer tulsk_…"

Codex


# ~/.codex/config.toml
[mcp_servers.tulsk]
url = "https://tulsk.io/api/mcp/mcp"
http_headers = { Authorization = "Bearer tulsk_…" }

Cursor


// .cursor/mcp.json
{ "mcpServers": { "tulsk": { "url": "https://tulsk.io/api/mcp/mcp", "headers": { "Authorization": "Bearer tulsk_…" } } } }

Outside agents never approve requests, change agents, schedules or webhooks, or see billing, and never send email or Slack messages directly: a Tulsk agent they start works under its own approvals. Memory they write is marked as outside content; comments, and the history of tasks they create or update, show “via” the token’s name; other writes carry no mark, and every call appears under the token in Settings.