Arcade vs Toolhouse (2026)

More than 8,000 permission aware tools are available, plus custom or third party MCP servers that the customer builds or brings. Authorization is delegated, so the agent acts as its user within that user's scope, and Arcade handles OAuth across dozens of auth providers. Hosted remote MCP servers connect through Arcade to systems such as Salesforce, ServiceNow, GitHub, Atlassian and Snowflake.

Tool Recommendation, added in October 2026, lets an agent describe a task and get back the few tools that fit instead of loading the whole catalog, which Arcade puts at up to 91% fewer input tokens per session, and a Private Registry lets users browse the tools behind a gateway and request missing ones.

Sourcearcade.devread 2026-10-24

Workers connect to more than a thousand services through built in integrations Toolhouse maintains and through integrations that need a connection, authorized per service over OAuth through Auth Connections, with the actions a worker may perform chosen per integration. Workers also call tools on remote MCP servers over SSE or streamable HTTP with OAuth, and can call third party APIs from uploaded API documentation in the virtual computer.

SourceToolhouse, docs.toolhouse.ai toolhouse/integrations, auth-connections and mcp-serversread 2026-09-21

Arcade executes the tool calls an agent makes and returns the results. Sequencing, branching and handoffs between agents live in the customer's own framework. Setup guides cover frameworks such as LangChain, CrewAI, OpenAI Agents, Google ADK and Mastra, among others. No workflow runtime of Arcade's own is named, so retries, routing and deterministic workflow steps sit outside the platform.

Sourcearcade.devread 2026-10-24

A manager worker coordinates specialized workers through the Call Agent integration, deciding which to involve, running them in sequence, in parallel or mixed according to data dependencies, passing one worker's output to the next, and chaining managers to any depth. Workers can also trigger other workers as a pipeline. The customer configures that routing across steps.

SourceToolhouse, docs.toolhouse.ai capabilites/worker-orchestrationread 2026-09-21

Tools in Arcade run when an agent or MCP client calls them. No schedule, event, webhook or inbound queue that starts an agent run is named.

Sourcearcade.devread 2026-10-24

Schedules start a worker autonomously on a cadence set in natural language or in the interface, as often as every ten minutes and with several schedules per worker, and workers also start on events such as a webhook notification from another service, so work reaches the agent without a person initiating it. Each worker also has its own email inbox, a chat interface and an API endpoint, with an optional completion webhook.

SourceToolhouse, docs.toolhouse.ai toolhouse/schedules and introduction-to-ai-workersread 2026-09-21

Agents get authenticated tools that read from and act in outside systems. Arcade names no index, graph, embeddings layer or other retrieval structure that it maintains over the customer's documents.

Sourcearcade.devread 2026-10-24

Agent Files attach knowledge to a worker, such as guidelines, templates, skill files and reference documents in text or PDF, and their contents are added directly to the worker's instructions, where they persist for every run. The document parser and file download bring further files in for each task. There is no index, graph or embeddings layer over the customer's knowledge that scales past the context window and stays queryable.

SourceToolhouse, docs.toolhouse.ai capabilites/skills-and-knowledge and capabilites/document-parserread 2026-09-21

Each user's OAuth tokens and authorization grants are kept, so tools can act without asking for consent again. Those stored credentials carry no agent state. Arcade names no memory that keeps an agent's state or a user's context across runs.

Sourcearcade.devread 2026-10-24

Workers write memories of data, behavior and past outputs through a Memory Store integration and read the relevant ones back through Semantic Memory Search on later runs, with reading and writing separable so a worker can be read-only, and each run keeps its full conversation history under a run ID across chat, email and the API. Toolhouse states no lifetime for stored memories and offers no surface to review, edit or delete them.

SourceToolhouse, docs.toolhouse.ai capabilites/memory and developers/worker-statefulnessread 2026-09-21

Contextual access runs the customer's own policy logic inline. An access hook decides which tools each user can see. A pre execution hook allows, denies or modifies inputs before each tool call, and a post execution hook allows, denies or modifies the output, so sensitive actions can be blocked and personal data redacted.

Users authorize each scope before a tool acts for them, an agent can line up every authorization a job needs before it starts, and rate limits apply. A tool author can ask the user for input during a call through MCP elicitation, but Arcade names no step where a person reviews or approves an action before it commits.

Sourcedocs.arcade.dev/en/operate/governance/contextual-accessread 2026-10-24

Agent Builder writes a governance section into each worker's instructions stating what controls apply and when it should stop and escalate, for example drafting a campaign and requiring approval before sending, and a worker can start autonomously and email its owner for approval before continuing. The approval lives in the worker's prompt, not in a gate the platform enforces before an action runs, so the boundary is stated, not enforced.

SourceToolhouse, docs.toolhouse.ai introduction-to-ai-workers and toolhouse/worker-emailsread 2026-09-21

Authentication runs against the customer's identity provider, with Okta, Microsoft Entra ID, Auth0 and Stytch as user sources. Authorization is delegated, so an agent acts as its user and never past that user's scope. Contextual access hooks enforce the customer's policies on every tool call. Role based access control has organization admin, project admin and member roles, and SSO and audit logs are included on Enterprise. No security attestation is published.

Sourcearcade.devread 2026-10-24

Workers reach only the services and actions they are given, through OAuth auth connections scoped per worker and action selection per integration, a least privilege model for tool access. A worker can be kept private to its owner, API keys are encrypted in transit and at rest, customer data is encrypted, not used for model training and scoped to the workspace, and Enterprise customers can arrange custom authentication models. There is no identity provider integration such as SSO or SAML and no attestation or trust center, and the toolhouse.ai footer links only privacy and terms pages.

SourceToolhouse, docs.toolhouse.ai auth-connections, developers/authentication and toolhouse.ai/pricingread 2026-09-21

Every action taken through Arcade leaves one record naming the agent, the user and the system, streamed to the customer's SIEM, and execution tool logs show which tools ran, who they ran for and whether they succeeded. Administrative actions go to a separate audit log that is on by default, with OpenTelemetry export, a filterable dashboard and a REST API for pulling events into the customer's SIEM or compliance tooling.

Sourcedocs.arcade.dev/en/operate/governance/audit-logsread 2026-10-24

The Logs page records every worker execution with status, run and user IDs and duration, a trace of each integration call in order with a timing waterfall and its exact input and output, and a history of the worker's intermediate outputs and any approvals or directions a person gave, with log retention stated by plan: 30 days on Business, one year on Business Max and custom on Enterprise.

SourceToolhouse, docs.toolhouse.ai toolhouse/logs and toolhouse.ai/pricingread 2026-09-21

Customers choose between Arcade Cloud and their own infrastructure. A turnkey platform that Arcade operates can be deployed into the customer's cloud account, as an Azure Marketplace managed app or through an AWS private offer. A self managed Helm deployment can run on the customer's Kubernetes cluster, for data residency, network isolation or compliance. Enterprise adds VPC and fully air gapped deployment.

Sourcedocs.arcade.dev/en/operate/deployread 2026-10-24

Toolhouse is a hosted service on every plan, Business, Business Max and Enterprise, with no self hosted, private cloud, on premises or region selection option.

SourceToolhouse, toolhouse.ai/pricing and docs.toolhouse.ai; toolhouse.ai/en/pricingread 2026-09-21

The catalog holds tools and MCP servers that give an agent actions to take. The sample apps are example code. Arcade names no ready made workflows, templates or prebuilt agents of its own.

Sourcearcade.devread 2026-10-24

Agent Builder assembles each worker from built in capabilities Toolhouse maintains, among them scraping, web and social search, memory, image generation, vision, a virtual computer and document parsing, and common worker patterns such as briefing, triage and content creation are ones the customer builds from a prompt. Workers an organization has built can be shared with its team. Toolhouse offers no catalog or named set of ready made workers a customer adopts, so these are assets the customer assembles.

SourceToolhouse, docs.toolhouse.ai readme and introduction-to-ai-workersread 2026-09-21

Any client, model or framework the customer picks can use Arcade's tools. Setup guides cover the customer's own LLM and frameworks such as LangChain, OpenAI Agents, Google ADK, CrewAI, Mastra and Vercel AI SDK, among others. The evaluation command runs against whichever providers and models the customer names.

Sourcearcade.devread 2026-10-24

Every plan includes model usage, which Toolhouse calls free unlimited tokens. The models or providers workers run on are not named, the customer cannot choose one, and nothing describes working across models, routing between them, or bringing one's own provider and keys.

SourceToolhouse, toolhouse.ai/pricing and docs.toolhouse.ai; toolhouse.ai/en/pricingread 2026-09-21

Developers get an API, described in the Arcade API Reference, and a Contextual Access Webhook API. SDKs appear throughout the framework guides, alongside the Arcade CLI and an open source MCP framework in Python and JavaScript for building MCP servers that Arcade hosts or connects. The Arcade plugin, released in October 2026, installs into Claude Code, Cursor, Codex, VS Code, Copilot CLI and Claude Desktop, and any remote MCP client can connect through one server URL. The REST API behind audit logs is called with an Arcade API key.

Sourcedocs.arcade.dev/llms.txtread 2026-10-24

Every worker is callable as an API at its own endpoint, authenticated with a Toolhouse API key. POST starts a task, PUT continues it under a run ID, GET returns the full history, NDJSON endpoints stream tool calls alongside the output, and long virtual computer jobs run in the background for polling.

SourceToolhouse, docs.toolhouse.ai developers/workers-api and developers/authenticationread 2026-09-21

The arcade evals command runs evaluation suites that the customer writes against its tools. It gives detailed results and critic feedback, a failures only view, multiple runs for stability and comparisons across providers and models in one run. Capture mode records the tool calls a model makes without scoring them, so they can become test expectations, and Arcade describes its evals as catching hallucinated and destructive calls before they run.

Sourcedocs.arcade.dev/en/build/create-tools/evaluate-tools/run-evaluationsread 2026-10-24

The Testing Workbench runs a new or edited worker on real test inputs so its builder can check the output before use and correct the worker through Agent Builder. There are no scored test cases, datasets, evaluation harness or quality gate that returns a verdict.

SourceToolhouse, docs.toolhouse.ai toolhouse/quickstart and introduction-to-ai-workersread 2026-09-21

Actions go through authenticated APIs and MCP tools. Arcade names no browser, desktop or remote computer session that it runs for an agent. Some toolkits connect to scraping and sandbox services that other vendors run.

Sourcearcade.devread 2026-10-24

A built in scraper that Toolhouse maintains, attached when a task needs it, reads web pages at run time with location selection by country, and Vision can capture and read a page screenshot. Toolhouse says the scraper reads pages and does not fill forms or log in, which is web scraping, not an agent operating a browser. The virtual computer executes Python code and operates no interface.

SourceToolhouse, docs.toolhouse.ai capabilites/scraping and capabilites/visionread 2026-09-21