feat(agent): Add in-process MCP support

[jordan-umusu]

Summary by cubic

Adds in-process MCP support so agents can run command-based MCP servers (stdio) inside the sandbox. Extends MCP integrations, presets, runtime, and UI to configure, validate, and run these servers securely.

• New Features

  • Added MCP server_type: url or command, with new fields (command, command_args, command_env, timeout).
  • Frontend integration dialog supports server type selection, validation, allowed commands (npx, uvx, python, python3, node), args (one per line), env (JSON), and timeout.
  • Integration detail page supports command-type configs (edit via dialog, delete with confirmation).
  • API and schemas updated: server_uri is optional for command type; command_env is write-only (exposed as has_command_env); workspace_id schema fixed to uuid.
  • Presets resolve MCP integrations into URL and command servers, re-validate command configs, and pass mcp_command_servers through AgentConfig.
  • Loopback and Claude runtime consume mcp_command_servers, start subprocess servers, and surface stderr as stream error events; avoid name collisions by suffixing server names (-cmd) if needed.
  • Security validation for command servers to prevent injection (allowlisted commands, safe args/env).
  • Sandbox image adds node, npm/npx, and uvx for running MCP servers.

• Migration

  • Run Alembic migration to add server_type and command-related fields, and make server_uri nullable.
  • Rebuild the Docker image to include Node and uvx tooling.
  • For command-type MCP servers, enable internet access in the preset (agents need network access inside the sandbox).
  • Existing URL-based integrations continue to work; server_type defaults to url.

^{Written for commit a9268b0. Summary will update on new commits.}

[jordan-umusu]

@cubic review

[jordan-umusu]

@codex review

[cubic-dev-ai]

@cubic review

@jordan-umusu I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

1 issue found across 21 files

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="frontend/src/components/integrations/mcp-integration-dialog.tsx">

<violation number="1" location="frontend/src/components/integrations/mcp-integration-dialog.tsx:337">
P2: Guard command_env parsing by server_type. As written, JSON.parse runs even for URL-type integrations, so stale/invalid command_env input can throw and block submit after switching server type.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f73368e620

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[cubic-dev-ai]

1 issue found across 21 files

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="frontend/src/components/integrations/mcp-integration-dialog.tsx">

<violation number="1" location="frontend/src/components/integrations/mcp-integration-dialog.tsx:197">
P2: Command env validation allows non-string values, but the API expects `Record<string, string>`. Validate that all values are strings before accepting the JSON.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 08b85f3662

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".