Taskade

The vendor states agents carry 22 or more built-in tools plus custom tool definitions, with 34 tools cited in later material, and Automations connect more than 100 native integrations with branching, looping and filtering.

Named integrations span Slack, Gmail, HubSpot, Salesforce and Stripe, with further reach through Zapier and n8n, for which the vendor maintains a public source-of-truth repository of Taskade actions and triggers built on its public API. Documented agent actions include creating tasks, sending emails, posting to Slack and updating calendars. The official MCP server exposes a further 50-plus tools for workspaces, projects, tasks, agents, knowledge bases, templates and automations.

Sourcegithub.com/taskade/mcp, github.com/taskade organization repositories, github.com/taskade/taskade READMEread 2026-08-31

The vendor states AI Agents v2 implements multi-agent teams alongside custom tools, persistent memory, slash commands and human-in-the-loop controls, and describes agents deployed as multi-agent teams with a central orchestrator coordinating agent activity and the app lifecycle.

Automations are documented as no-code workflow automation with branching, looping and filtering across more than 100 integrations. Workspace DNA describes memory feeding agents, agents triggering automations, and execution writing new state back. Automations and agents are addressable objects in the MCP server tool surface.

Sourcegithub.com/taskade/taskade README, github.com/taskade/mcp, taskade.com AI agents pageread 2026-08-31

The vendor's documentation index describes Agent Knowledge as training agents on uploaded PDFs, CSVs and documents, URLs and YouTube, or connected projects as live, auto-updating sources, with a Knowledge Management page covering bulk import and auto-categorization and REST API v1 endpoints to add and remove projects and media from an agent's knowledge. Each agent keeps its own maintained set of knowledge sources that update automatically. Long-Term Memory is a separate feature.

Sourcedocs.taskade.com documentation index, Agent Knowledge and Knowledge Management entries and Agents API referenceread 2026-09-29

The vendor's Tools for AI Agents page documents Human in the Loop as an approve or reject gate before agents interact with external platforms: the builder enables it in the agent's Tools tab and selects which tools require approval, and when the agent wants to use one of those tools it proposes the action, the user sees a confirmation prompt, and the agent proceeds only on Approve and stops on Reject.

Recommended uses are sending emails, posting to social media and creating CRM records. That is an approval step the vendor ships in the run path. Tools can also be enabled, disabled or removed per agent.

Sourcedocs.taskade.com Tools for AI Agents pageread 2026-09-29

Documented access controls are OAuth 2.0 and Personal API Token authentication for programmatic access, OIDC sign-in for Genesis apps, single sign-on on higher tiers, team workspaces with administrative controls, and role assignment within workspaces. Agent public access is a documented toggle controlling whether an agent is reachable by shareable link or embedded. No security certification, trust center, SCIM provisioning or audit-log documentation appears on the vendor's developer documentation, MCP repositories or product pages cited.

Sourcehelp.taskade.com developer API article, github.com/taskade/taskade README, github.com/taskade/mcp, taskade.com product pagesread 2026-08-31

Agent conversations are documented as auto-saved and persistent for later review, and automations are described as durable with execution resuming after failures. Advanced analytics are cited by the vendor on higher tiers. No agent execution trace, tool-call record, cross-agent run history, change audit log or cost attribution appears in the vendor's README, MCP tool surface, developer documentation or product pages cited. The Workspace DNA architecture has automation execution writing new state back into projects, so run-level attribution is consequential and is not documented.

Sourcegithub.com/taskade/taskade README, github.com/taskade/mcp, help.taskade.com developer API articleread 2026-08-31

The vendor's Long-Term Memory page documents memory held as a dedicated Memory Project that the agent writes and reads: EVE's workspace memory persists across sessions, is scoped to one workspace and grows automatically as the user interacts, and per-agent memory persists across conversations with that agent.

The user can browse the memory project, remove entries, correct facts, add context and version it, with edits taking effect in the next conversation, and the memory is addressable through the Action API v2. The agent writes this memory, it has a set scope, and users can purge, review and edit it; the memory project is separate from business projects, so deleting it does not delete business records.

Sourcedocs.taskade.com Long-Term Memory pageread 2026-09-29

Taskade is documented as a cloud workspace with native desktop and mobile applications and self-serve pricing starting at a low monthly rate. No self-hosting, on-premises deployment, virtual private cloud option, customer-managed encryption keys or customer-selectable region appears in the vendor's README, developer documentation, MCP repositories or product pages cited. Genesis apps can be published to a customer's own domain with OIDC sign-in, which controls the published application's address rather than where Taskade processes or stores data.

Sourcegithub.com/taskade/docs, github.com/taskade/taskade README, help.taskade.com developer API articleread 2026-08-31

The vendor's documentation directs users not to start from a blank prompt but to clone a ready-made app for free from the community gallery or the vendor's official kits and customize it in Taskade Genesis. A clone-ready App Kits Gallery is maintained by the vendor, and a Workspace DNA template for building Genesis apps is published as its own repository.

The template library is documented at 700-plus templates covering project management, engineering and marketing, and the vendor cites 150,000-plus apps generated. Templates are an addressable object type in the MCP server tool surface, with a documented flow for creating a project from a template.

Sourcegithub.com/taskade/docs, github.com/taskade organization repositories, github.com/taskade/mcpread 2026-08-31

The vendor's documentation index describes Automations as a trigger-and-action engine running work around the clock across more than 100 integrations with AI agents running the workflows, a trigger reference spanning schedules, Slack, Gmail, Google Sheets, HubSpot and GitHub events, and a Webhooks page on triggering automations with inbound webhooks; the July 2026 changelog records an expanded trigger catalog and fixed HTTP webhook triggers. An Autonomous Agents page covers agents that run without per-request prompting. Scheduled and event triggers wake agents without a user.

Sourcedocs.taskade.com documentation index, Automations, Comprehensive Integrations, Webhooks and Autonomous Agents entriesread 2026-09-29

The vendor's TSK-1 page documents a model picker in which the customer sets an agent or automation to Auto, pins one of four capability tiers (Instant, Standard, Thinking, Reasoning), or on Max and Enterprise pins a custom model such as a specific Claude or GPT version, which bypasses TSK-1 routing entirely. The Long-Term Memory page states more than eleven frontier models are integrated and that switching an agent from Claude to Gemini keeps its memory. Customers choose across model makers, though custom models are limited to the Max and Enterprise plans.

Sourcedocs.taskade.com TSK-1 Model Tiers and Long-Term Memory pagesread 2026-09-29

The vendor documents REST API v1, an Action API v2, an SDK and webhooks for programmatic access, with authentication by Personal API Token or OAuth 2.0 and OIDC sign-in for Genesis apps.

An official Model Context Protocol server is published as its own repository under the vendor's GitHub organization, exposing 50-plus tools covering workspaces, projects, tasks, agents, knowledge bases, templates and automations, alongside an OpenAPI-to-MCP code generator, and a hosted MCP endpoint at the vendor's domain connectable by OAuth without local installation. Agents can be published with a shareable link or embedded. Documentation is maintained as a public repository accepting pull requests.

Sourcegithub.com/taskade/mcp, github.com/taskade/docs, help.taskade.com developer API article, taskade.com MCP guidanceread 2026-08-31

No testing, evaluation, scoring, benchmarking or regression capability appears in the vendor's README, MCP server tool surface, developer documentation or product pages, and no self-checking or verification gate on agent output before use is described. Durable execution allowing automations to resume after failures is documented, which is reliability rather than measurement of output quality. The MCP tool surface covers workspaces, projects, tasks, agents, knowledge bases, templates and automations, with no evaluation or test object type.

Sourcegithub.com/taskade/taskade README, github.com/taskade/mcp, github.com/taskade/docs, help.taskade.com developer API articleread 2026-08-31

The vendor's own materials describe agents with 22 to 34 built-in tools, more than 100 native integrations, custom tool definitions and web browsing to gather information, with all action paths running through connectors, the REST and Action APIs, or the MCP server, which are programmatic interfaces. Web browsing for research is retrieval over fetched content rather than operation of software lacking an interface. No browser control, session automation, form filling or cloud-computer capability appears in the vendor's README, MCP tool surface, developer documentation or product pages.

Sourcegithub.com/taskade/taskade README, github.com/taskade/mcp, help.taskade.com developer API articleread 2026-08-31