Bezalel compared with Composio and Executor
Bezalel, Composio, and Executor all let AI agents use your apps over MCP.[2][11] For Composio and Executor, this page uses only what each company states on its public pages, checked on November 3, 2026. Each bracketed number links to the page a claim comes from, and those pages may have changed since.
Bezalel's app connectors run on Composio's app catalog. When an agent uses an app from that catalog through Bezalel, Bezalel passes the call to Composio.
Composio offers more than one MCP product. When a claim is about one of them, such as Composio Connect, its MCP server for existing MCP clients,[1] the table names it. Executor's current docs describe Executor 2, which is in beta.[10]
Side by side
| What it is | |
|---|---|
| Bezalel | One MCP server that gives your agents your connected apps, a shared memory, and other tools. |
| Composio | Connects AI agents to external apps through tools, managed authentication, and execution.[1] |
| Executor | Connects the tools you use to the agent you already use, through one MCP server. Your agent can build tools, automations, and interfaces on top.[9] |
| How agents connect | |
| Bezalel | One MCP endpoint over Streamable HTTP. Each agent gets its own bearer token. |
| Composio | Composio Connect is one MCP server over Streamable HTTP. Clients sign in with browser OAuth or a consumer API key.[1] |
| Executor | One MCP endpoint at /mcp over Streamable HTTP.[11] Clients use browser sign-in, a personal access token, or a local server key.[12] |
| Apps | |
| Bezalel | Hundreds of apps from Composio's catalog, such as GitHub, Slack, Gmail, and Notion. Connect each once, and every agent you allow can use it. |
| Composio | 1,000+ apps through Composio Connect, per its docs.[2] You can reuse connections across different AI assistants.[3] You can bring your own tools and MCP servers alongside the catalog.[7] |
| Executor | Apps such as Gmail, Calendar, GitHub, Linear, and Stripe, plus MCP servers, OpenAPI, and GraphQL APIs you add by URL. Your agent can also write new tools.[9][13] Every account you connect is available to every agent you use.[9] |
| Shared memory | |
| Bezalel | One memory shared by every agent you give memory access. Browse or delete any entry from the dashboard. |
| Composio | Not described on its public pages. |
| Executor | Not described on its public pages. |
| Per-agent permissions | |
| Bezalel | You turn capabilities on for each agent, such as connectors or memory. You set allowed app actions per app, for the whole account. |
| Composio | You can limit which actions your AI can take in an app and require approval for some.[3] Its docs mark approvals as experimental.[8] Its MCP Gateway for enterprises manages access by organization, team, user, agent, or toolkit.[4] |
| Executor | Connect an agent through a scoped connection, a named MCP URL that reaches only the apps, accounts, and tools you pick.[11] Tools can require a person's approval.[14] |
| Activity records | |
| Bezalel | The Activity page lists each agent's app actions, emails, memory changes, texts, and computer and sandbox use, with the time and whether it worked. Its activity records leave out app action arguments and results. App actions run through Composio, which keeps its own logs.[5] |
| Composio | Logs each tool execution, including its arguments and results by default, for up to one year. Zero data retention, which keeps only metadata, is available on paid plans and included on Enterprise.[5] |
| Executor | Its pricing page lists audit logs for every tool call on the Enterprise plan.[15] Local, desktop, and self-hosted builds keep up to seven days of traces by default, within a 1 GiB size target.[16] |
| Other capabilities | |
| Bezalel | Email inboxes, iMessage and Telegram texting, card purchases you approve, a cloud computer, and code sandboxes. |
| Composio | Triggers that deliver app events to your webhook.[7] A browser for sites without APIs and a sandbox to run code.[6] |
| Executor | Automations your agent writes, run on a schedule or when a webhook arrives. Its homepage marks a browser and a computer as coming soon.[9] |
| Hosting and license | |
| Bezalel | Hosted at bezalel.sh. |
| Composio | Composio Cloud, or a private VPC or self-hosted deployment arranged for enterprise needs.[6] |
| Executor | Executor Cloud, a desktop app (beta), a local server run from the CLI, or self-hosted on infrastructure you control.[9] The CLI, desktop app, and self-hosted server are MIT licensed and free.[17] |
| Pricing | |
| Bezalel | One owner per account, with many agents. See Bezalel's pricing page. |
| Composio | Free Hobby plan, a paid Pro plan, and custom Enterprise pricing. Composio meters tool calls.[7] |
| Executor | Cloud is free for up to 3 members. Team costs a flat price per member each month.[15] |
When each one fits
Which one fits depends on what you need besides the apps. If you only need apps in an MCP client, Composio Connect gives your agent 1,000+ apps through one connection.[2] Composio's session SDKs are for applications serving their own users.[1] Executor's homepage says you can bring your own integrations, change how any tool works, and build apps on top of them,[9] and Executor 2 is in beta.[10] Bezalel is for running several agents that share the same apps and one memory, each with its own token and permissions, with email, texting, purchases you approve, a computer, and sandboxes over the same connection.
Sources
- Composio llms.txt: composio.dev/llms.txt
- Composio Connect docs: docs.composio.dev/docs/composio-connect
- Composio homepage: composio.dev
- Composio MCP Gateway page: composio.dev/mcp-gateway
- Composio data retention docs: docs.composio.dev/docs/security/data-retention
- Composio Enterprise page: composio.dev/enterprise
- Composio pricing page: composio.dev/pricing
- Composio approvals docs: docs.composio.dev/docs/approvals
- Executor homepage: executor.sh
- Executor command line docs: executor.sh/docs/run/cli
- Executor MCP endpoint docs: executor.sh/docs/mcp
- Executor MCP client docs: executor.sh/docs/mcp-clients
- Executor app authoring docs: executor.sh/docs/author-an-app
- Executor tools and approvals docs: executor.sh/docs/concepts/tools-and-approvals
- Executor pricing page: executor.sh/pricing
- Executor tracing docs: executor.sh/docs/run/tracing
- Executor pricing page, markdown version: executor.sh/pricing.md