WorkflowStack AI
WorkflowsIndustriesToolsGuidesAI QuizBlogEnterprise
Get Free Workflows
WorkflowStack AI

Practical AI workflows for SMB operators and enterprise teams. No fluff. No hype. Just what ships.

Library

  • All Workflows
  • Industries
  • Enterprise
  • Tools
  • Guides

Company

  • About
  • Blog
  • Newsletter
  • Contact

Stay Updated

Weekly workflow ideas for operators and enterprise teams.

Get Free Workflows

© 2026 Blueteem LLC. All rights reserved.

Privacy PolicyTerms of Service
HomeBlogComposio vs Arcade vs MCP in 2026: Agent Tool Integration Pricing, Auth and Security
November 10, 2026

Composio vs Arcade vs MCP in 2026: Agent Tool Integration Pricing, Auth and Security

The managed tool layers for AI agents price the same action four different ways. Here is what Composio, Arcade.dev and Pipedream Connect publish, what running your own MCP servers costs you instead, and a worked monthly bill.

The short answer

Composio is the cheapest managed option at list price: its free Hobby plan includes 100,000 tool calls a month, and Pro overage is $0.0003 per call. Arcade.dev charges $25 a month plus $0.01 per tool call, with VPC or air-gapped deployment on custom Enterprise. Self-hosted MCP has no licence fee, but you own OAuth, tokens and security. At 50,000 calls a month on your own OAuth apps, Composio Pro costs $29 and Arcade Team costs $525.

What these products actually are

An agent that sends email, updates a CRM or reads a calendar needs three things for each end user: an OAuth grant, a stored and refreshed token, and a tool definition the model can call. A managed tool layer provides all three. The Model Context Protocol (MCP) is the open standard for the last part, the way an agent discovers and invokes tools. It does not run anything for you. See the MCP glossary entry for the basics.

Composio advertises 1,700+ app toolkits. Its pricing page says it handles "Consent, token storage, refresh, and scopes ... for every connected account."

Arcade.dev calls itself "the actions runtime for production AI agents". It says "Agents act on behalf of the authenticated user, never through broad service accounts."

Pipedream Connect advertises "Managed auth for 3,000+ APIs" and "10,000+ prebuilt tools & triggers".

Self-hosted MCP means you write or deploy MCP servers yourself, on your own infrastructure. The specification is open; per the repository's LICENSE, new contributions are Apache-2.0 while the project transitions from MIT.

Both Composio and Arcade sit in our AI Infrastructure category. If you are shopping beyond these two, see Composio alternatives.

Published prices

Prices and plan names are as published on each vendor's own pricing page; last verified 10 November 2026.

Published prices
VendorBilling unitPublished entry priceWhat it's for
Composio (developer platform)Tool calls, trigger eventsHobby $0 with 100,000 tool calls a month, hard-capped. Pro $29 a month including $29 of credits, then $0.0003 per tool callSDK, managed OAuth and MCP for apps you ship to your own users
Arcade.devTool callsFree: 2,000 tool calls a month. Team: $25 a month plus $0.01 per tool callPer-user authorized tool execution; VPC or air-gapped deployment on Enterprise
Pipedream ConnectExternal users, plus usage creditsStartup $99 a month billed annually, or $150 billed monthly. Includes 100 users (then $2 each) and 10,000 credits (then $0.012 each)Managed auth and prebuilt actions inside your app
Self-hosted MCPYour own computeNo licence feeFull control, full responsibility

Composio's developer plans apply to signups on or after 15 August 2026; earlier customers stay on their existing plan through 31 December 2026. Composio publishes two separate price lists. The plans above are for its developer platform. Its "For You" plans ($18 per seat per month for Standard) cover connecting staff to tools such as Claude or Cursor, which is a different product. Smithery is no longer an option. It shut down its MCP proxy and managed identity services on 31 October 2026, and smithery.ai now says "Smithery is now part of Arcade" and remains only as "a public (non-proxied) MCP registry". Klavis's pricing URL now describes "live environments for training AI agents", with no list prices.

The fine print in each pricing unit

Composio: "Each tool execution counts once. Meta tools such as tool search are free." The 100,000 free calls assume your own OAuth app. On Composio-managed apps, you get 20,000 free a month, then "Hobby ... stops" and Pro pays "by app". Composio says managed apps are "not built for production scale". Add-ons are priced per call: running a tool outside a Session is +$0.0001, a BAA is +$0.0003, and zero data retention is +$0.0001. Hobby allows 3 team members and 2,000 API requests per minute; Pro allows 10,000.

Arcade: the Team plan lists "Tool calls at $0.01 each", and the page does not mention any included calls on Team. Enterprise "includes annual bundles of tool calls" with "discounted additional tool call rates", but no figures are published.

Pipedream: a credit is "1 credit per 30 seconds of compute" for actions, MCP tool calls and proxy requests, so a slow tool can cost more than one credit. An external user is "each distinct user of your app who connects at least one account". Serving real users requires Startup; Free is development-only.

Worked example: one agent, 200 end users

Assumptions: a B2B product embeds one agent. 200 end users each connect two apps, making 400 connected accounts. Each user triggers 250 tool calls a month, for 50,000 calls in total. Every call finishes in under 30 seconds, you use your own OAuth apps, calls run inside Composio Sessions, and billing is monthly.

Worked example: one agent, 200 end users
VendorArithmeticMonthly total
Composio Hobby50,000 is within the 100,000 free calls; connected accounts are "Unlimited & free"$0
Composio Pro (for 10,000 req/min and no hard cap)$29 plan; usage within free allowance$29
Arcade Team$25 + (50,000 × $0.01)$525
Pipedream Startup, monthly$150 + (100 extra users × $2) + ((50,000 − 10,000) × $0.012)$830
Self-hosted MCPYour hosting plus engineering timeNot a list price

At 10 times the volume (2,000 users, 500,000 calls), list prices diverge further. Composio Pro meters (500,000 − 100,000) × $0.0003 = $120. Its page says Pro "includes everything in Hobby plus $29 in monthly credits", so the bill is about $120 if the included credit offsets usage; confirm with Composio. Arcade Team is $25 + $5,000 = $5,025. Pipedream Startup is $150 + (1,900 × $2) + (490,000 × $0.012) = $9,830 at the published credit rate; Pipedream says that rate "drops at higher volume" but publishes no tiers. Arcade's Enterprise discounts are also unpublished.

These numbers cover tool execution only. Model tokens, and any per-app rates on Composio-managed apps, come on top.

Get one new AI workflow every week

Practical playbooks, prompts, and tool stacks — for the SMB operators and enterprise teams actually shipping AI.

Auth: per-user OAuth, token storage and scopes

All three managed vendors store and refresh tokens per end user, so your app holds an identifier rather than credentials.

Composio: "User authentication requires a userID", and "Composio automatically refreshes OAuth tokens before they expire." You can instead "Pass your own access tokens at execution time. Composio never stores or refreshes them." For production, its docs say to generate your own OAuth credentials.

Arcade: "Each tool in Arcade's MCP servers has a set of required permissions", and the Engine "will automatically prompt the user to authorize the tool". Its docs describe "AES-256 encrypted storage volumes", with additional AES-256 encryption on tokens and secrets. Its provider docs (Attio, for example) say production use needs your own app credentials.

Pipedream: "Credentials are encrypted at rest, scoped per project, and never exposed to your app."

Self-hosted MCP: you build the token store yourself, and the spec puts hard rules on it. The MCP authorization spec says servers "MUST validate that access tokens were issued specifically for them as the intended audience". The security guide says: "MCP servers MUST NOT accept any tokens that were not explicitly issued for the MCP server." Forwarding a user's token straight to a downstream API, which the spec calls token passthrough, is forbidden.

Compliance and self-hosting, as stated by vendors

Compliance and self-hosting, as stated by vendors
VendorAttestation statedSelf-hosting
Composio"SOC 2 Type II attested and ISO 27001:2022 certified"Homepage lists "Composio Cloud", "Your own VPC" and "Fully self-hosted"; confirm which apply to the developer platform
Arcade.dev"SOC 2 compliant" (homepage)"Cloud, on-prem, air-gapped, or hybrid". Self-host with Helm
Pipedream"SOC 2 Type 2 and GDPR" on every plan. HIPAA on BusinessNot stated on the pricing page
Self-hosted MCPYour ownYes, by definition

Composio offers a HIPAA BAA as a Pro add-on and includes it on Enterprise. Ask each vendor whether its SOC 2 scope covers the deployment model you will actually use.

The security risks of MCP servers

The MCP specification (version 2026-07-28) publishes its own threat list. These are the risks that matter most when choosing between managed and self-hosted:

  • Confused deputy: an MCP proxy using a static OAuth client ID can be tricked into skipping user consent. The spec says such proxies "MUST implement per-client consent".
  • Token passthrough: forbidden, as covered above.
  • Server-side request forgery: a malicious server can point OAuth discovery URLs at internal addresses, including 169.254.169.254 cloud metadata.
  • Local server compromise: a one-click local install runs a command on the user's machine. Clients "MUST" show "the exact command that will be executed, without truncation".
  • Scope inflation: the spec warns against wildcard scopes and recommends a "progressive, least-privilege scope model".

The tools section adds two rules that apply whoever hosts the server: "there SHOULD always be a human in the loop with the ability to deny tool invocations", and "clients MUST consider tool annotations to be untrusted unless they come from trusted servers." Neither rule stops a tool result from carrying injected instructions. Composio sells a Firewall that it says "blocks instructions hidden in tool results before your agent reads them", priced at $0.50 of credits per 1M tokens on Pro. That is a vendor claim, not an independent test. See human-in-the-loop for approval patterns.

When not to use each

Composio: avoid it if you will stay on Composio-managed OAuth apps at scale. The free allowance on those is 20,000 calls, and Composio itself says they are "not built for production scale". Hobby's hard cap also pauses your agent mid-month, so production needs Pro.

Arcade.dev: avoid it for high-volume, low-value calls. At $0.01 per call, it costs 33 times Composio's Pro overage rate. Avoid it too if you need volume pricing without an Enterprise contract.

Pipedream Connect: avoid it if most of your users make few calls. The $2 per external user beyond 100 is the dominant cost at 2,000 users.

Any managed tool layer: plan an exit path. Smithery shut down its connection-management services on 31 October 2026.

Self-hosted MCP: avoid it if no one on the team will own OAuth consent screens, audience validation and SSRF protection. Every MUST in the spec becomes your audit finding.

Frequently asked questions

Is Composio cheaper than Arcade?

At list price, yes. At 50,000 tool calls a month on your own OAuth apps, Composio Pro costs $29 (its free Hobby plan covers the volume but hard-stops at its cap), while Arcade Team costs $25 plus $0.01 per call, which is $525.

How much does Arcade.dev cost?

Arcade.dev has a free plan with 2,000 tool calls a month. The Team plan is $25 a month plus $0.01 per tool call. Enterprise pricing is custom and includes annual bundles of tool calls.

What is the difference between MCP and Composio?

MCP is an open protocol for how agents discover and call tools; it does not store tokens or host servers. Composio is a managed service that handles per-user OAuth, token refresh and 1,700+ toolkits, and it also exposes tools over MCP.

Do I need Composio or Arcade if I run my own MCP servers?

Not necessarily, but you take on per-user OAuth, token storage and the MCP spec's security requirements yourself. Those include rejecting tokens not issued for your server and implementing per-client consent on proxies.

Are MCP servers safe to use?

They are as safe as their implementation. The MCP specification lists confused-deputy, token-passthrough, SSRF and local-server-compromise risks, says clients must treat tool annotations from untrusted servers as untrusted, and recommends a human able to deny tool invocations.

Is Composio SOC 2 compliant?

Composio states it is SOC 2 Type II attested and ISO 27001:2022 certified. A HIPAA BAA is a paid add-on on Pro and included on Enterprise.

Next step

Before agents get production credentials, put every one of them on an agent inventory and governance program. If you are still choosing an agent framework, compare LangGraph against the rest of the agent development tools.

Related Workflows

Internal Developer Productivity Agent Platform

A governed internal platform for shipping coding and ops agents to engineering teams — with shared guardrails, evals, and observability instead of shadow tools.

View workflow

AI-Powered Price Quoting

Generate accurate price quotes in minutes instead of hours by using AI to analyze job details and pull from your pricing database.

View workflow

AI Bookkeeping Categorization Agent

Auto-categorizes daily transactions, flags anomalies, and drafts a monthly P&L narrative — your bookkeeper reviews, doesn't reclassify.

View workflow

Keep Reading

March 19, 2026

Vapi vs Bland vs Retell (2026): AI Voice Agent Pricing, Limits and Real Monthly Cost

Vapi, Retell and Bland price AI phone agents in three different ways. We read each vendor's own pricing page and worked out what 2,000 inbound minutes a month actually costs on each.

November 10, 2026

Langfuse vs LangSmith vs Braintrust (and PostHog): LLM Observability Pricing 2026

The six LLM observability tools buyers compare most, priced from each vendor's own pricing page, with a worked monthly bill at 300,000 traces and a five-person team.

September 27, 2026

Fin vs. Zendesk vs. Gorgias vs. Freshdesk: The AI Support Agent Is Now Priced Four Different Ways

Fin bills an outcome, Zendesk a resolution, Gorgias an interaction, Freshdesk a 72-hour session, HubSpot a resolved conversation. Same 3,000 tickets, five invoices that differ by more than 3x.

Found this helpful?

Get weekly AI workflow ideas in your inbox.