Last updated: 2026-10-29
Executor is an open source integration layer with 4.3k GitHub stars that connects AI agents such as Claude Code, Codex and Cursor to services like Gmail, GitHub, Linear and Stripe through one MCP server. Agents can also write custom tools, schedules, webhooks and workflows on top. Executor itself runs no AI model.
About Executor
The product comes from a company backed by Y Combinator and led by founder Rhys Sullivan, and it sits between your AI agents and the accounts they need to reach. You add Executor to an agent once as a single MCP server, then connect accounts such as Gmail, Calendar, GitHub, Linear, Slack, Notion, Stripe, Drive, PostHog, Sentry and X through a secure link, so credentials stay out of chat and out of code. Every account you connect is then available to every agent you use, including Claude Code, Codex, Cursor, ChatGPT and Gemini.
The vendor describes Executor as the cloud for agent apps. An app can be a single tool or a bundle of tools, skills, storage in its own SQLite database, a page for people, schedules, webhooks and durable workflows. Tools can be sourced from MCP servers, OpenAPI specifications or GraphQL schemas, and they are ordinary JavaScript. Your agent writes the files and Executor builds, hosts and versions them. Executor runs no AI model of its own, so it extends the agent you already pay for rather than replacing it.
The vendor's own examples show the shape of the work: copy new Stripe customers into Notion on a timer, post a weekly GitHub summary to Slack, open a Linear ticket when a Sentry issue spikes, or hold outgoing Gmail for a few minutes so a send can be cancelled. It fits developers who already work through a coding agent. Against Zapier, n8n and Make there is no flow builder: the agent writes the automation as code and Executor runs it on a schedule or when a webhook arrives. To compare gateways with agent platforms more broadly, see best AI agent builders and best personal AI agents. If the agent you pair it with is still undecided, Cursor Composer 2 vs Claude Code covers that choice. Against Composio, the vendor says the difference is that you can bring your own integrations, change how any tool works and build apps on top. Other gateway-style options listed on HokAI include Orthogonal and QueryPie AIP.
Executor Cloud has a free plan, a per-member team plan and a custom enterprise plan; exact figures are in the pricing notes. The local command line tool and the server you host yourself carry an MIT licence and no fee. According to the vendor, the cloud does not yet host app pages or app data, so apps that need them run locally or self-hosted today, and browser and computer tooling is marked as coming soon.
Pricing
Free: $0 per month, capped at 3 members, with unlimited integrations. Team: $15 per member per month after a 14-day free trial, and adds verified domains plus join by team domain. Enterprise: custom pricing for self-hosted or dedicated cloud, SSO/SAML, SCIM, audit logs on every tool call, dedicated support and onboarding; security reviews, a DPA and SOC 2 are available on request.
The CLI, desktop app and self-hosted server are MIT licensed and free. Source: executor.sh/pricing.md, read 2026-10-29.
| Tier | Monthly price | What it includes |
|---|---|---|
| Free | Free | |
| Team | $15/mo | |
| Enterprise | Custom |
Key Features
- One MCP server for every agent: Add Executor to Claude Code, Codex, Cursor, ChatGPT or Gemini once and each connected account is available to all of them.
- Connect accounts once: Accounts connect through a secure link and credentials are stored separately from app source, so they never appear in chat or code.
- Tools from any source: Import an MCP server, an OpenAPI spec or a GraphQL schema, or have the agent write a tool as plain JavaScript.
- Schedules and webhooks: An app runs on a cron schedule or when a service calls its webhook, so no agent has to be online for it to work.
- Durable workflows and storage: Multi-step work resumes after an interruption, and each app keeps its state in its own SQLite database.
- Approvals and self-hosting: Apps can require approval before a tool runs, and the self-hosted server keeps private app pages on infrastructure you control.
Pros
- The core is MIT licensed with public source on GitHub, so a team can audit and self-host the server instead of trusting a closed gateway.
- One connection per account is shared by every agent, which removes the repeated OAuth setup of running Claude Code, Codex and Cursor side by side.
- Automations are code the agent writes, so a Friday GitHub-to-Slack summary behaves the same way each run without an agent online.
Cons
- The vendor says cloud app pages and hosted app data are still in development, so some app types only work locally or self-hosted today.
- There is no visual flow builder, so non-developers who expect a canvas to click through have nothing equivalent.
- A young product from a Y Combinator Summer 2026 company, with security reviews and a DPA offered on request to Enterprise buyers rather than published up front.
Data Handling
- Training-data policy
- No AI model is operated by Executor.
- Compliance
- SOC 2 and DPA on request (Enterprise plan)
Frequently Asked Questions
How much does Executor cost in 2026?
Executor Cloud has three plans. Free costs $0 and is limited to 3 members, with no cap on integrations. Team is $15 per member per month once the 14-day trial ends. Enterprise is quoted individually, and the self-hosted server, CLI and desktop app carry no licence fee.
Is Executor free to use?
Yes, in two ways. The cloud Free plan suits a small team, and the open source CLI, desktop app and self-hosted server cost nothing beyond your own hardware. Hosted app pages and app data are not finished on the cloud yet, so check the docs before planning around them.
What are the best alternatives to Executor?
Composio is the closest direct comparison for connecting accounts to agents, and the vendor says Executor adds bring-your-own integrations and apps on top. Choose Zapier, n8n or Make when you want a visual builder instead of agent-written code. Orthogonal is another gateway-style option listed on HokAI.
How does Executor compare to Zapier in 2026?
Zapier is built around a canvas where a person designs each automation. Executor has no builder: the agent produces code for the job, and the platform fires it on a timer or an incoming webhook. Pick Executor if your team already works through a coding agent, and Zapier if the people building automations do not.
How do you get started with Executor?
Sign in at app.executor.sh or install the CLI or desktop app, then add Executor to your agent as an MCP server. The vendor publishes a setup prompt that a coding agent can follow to connect over MCP and deploy a first app. Connect each account through the secure link once and every agent can use it.
Top Alternatives
- Zapier: Pick Executor if your team builds automations through a coding agent; pick Zapier if the builders need a visual canvas.
- n8n: Choose Executor when agents write the automation as code; choose n8n when people design workflows visually.
- Make: Pick Executor to give agents shared accounts and hosted tools; pick Make for a scenario builder run by non-developers.