Add Unstructured Transform MCP starter template

[SudSampath]

What changed

• Added a reusable MCP server template list to the Add MCP Server modal.
• Added Unstructured Transform as a starter template with the streamable HTTP endpoint and Authorization header placeholder.
• Clarified what Unstructured Transform does: converts PDFs, DOCX, PPTX, HTML, and images, including scanned/image-only pages via OCR, into clean markdown, JSON, HTML, or text, with RAG preprocessing for chunking, enrichment, and embedding.
• Documented the Unstructured setup flow in the MCP tools docs, including the required UNSTRUCTURED_API_KEY environment variable and allowlist domain.

Why

Unstructured Transform is exposed as an MCP server, so discoverability fits best as a first-class MCP starter configuration instead of a guessed native REST connector. Selecting the template fills the server name, URL, transport, timeout, and header values while still using Sim's existing connection test and MCP execution flow.

Validation

• pnpm dlx @biomejs/biome@2.0.0-beta.5 check --no-errors-on-unmatched --files-ignore-unknown=true ...
• git diff --check

Full workspace checks were not run in this fresh clone because Bun is not installed in the local shell and dependencies are not installed.

[vercel]

Someone is attempting to deploy a commit to the Sim Team on Vercel.

A member of the Team first needs to authorize it.

[SudSampath]

Quick follow-up / product context:

This PR is meant to be the small MCP-first step: it gets Unstructured Transform into the MCP add flow so a Sim user can pick it, add their UNSTRUCTURED_API_KEY, and use the hosted Transform MCP server in agents/workflows.

For folks less familiar with it: Transform converts documents like PDFs, DOCX, PPTX, HTML, and images, including scanned/image-only pages via OCR, into clean markdown, JSON, HTML, or text. It also supports the RAG prep work around that: chunking, enrichment, and embedding.

If we want Transform to show up in the main Integrations catalog too, I think there are two possible follow-ups:

• MCP-backed integration card — my preferred path if it fits Sim's product model. Keep execution backed by the hosted MCP server, add a normal catalog card/page with icon/category/tags/docs, and have Add to Sim open or prefill this MCP template.
• Native Sim block/tool — definitely possible, but heavier. We'd need REST/OpenAPI docs, auth details, file upload/reference behavior, exact request/response schemas, sample successful responses, supported operation choices, and Unstructured brand assets. I avoided that here because the public quickstart is MCP-first, and I didn't want to guess a Sim-native schema.

Happy to adapt this to whatever pattern the Sim team prefers. If MCP-backed catalog cards are acceptable, I can follow up with that PR.

[cursor]

PR Summary

Low Risk
UI and documentation only; existing save, domain allowlist, and connection-test paths are reused with no changes to MCP execution or auth.

Overview
The Add MCP Server modal now shows Starter templates in add mode. Picking a template pre-fills name, transport, URL, timeout, and headers (with an empty header row), switches back to the form view, and clears connection-test / submit state.

A new mcp-server-templates module defines a typed template list; Unstructured Transform is the first entry (streamable-http, https://mcp.transform.unstructured.io, Authorization: Bearer {{UNSTRUCTURED_API_KEY}}).

The MCP tools docs add a Starter Configurations section covering the Unstructured setup (UNSTRUCTURED_API_KEY, pre-filled URL/headers) and noting mcp.transform.unstructured.io for ALLOWED_MCP_DOMAINS on self-hosted deployments.

^{Reviewed by Cursor Bugbot for commit b9e441d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

This PR adds an MCP starter template for Unstructured Transform. The main changes are:

• A reusable starter template list for the Add MCP Server modal.
• A new Unstructured Transform template with URL, timeout, transport, and Authorization header defaults.
• Docs for the Unstructured setup flow and allowlist domain.

Confidence Score: 4/5

The new external MCP starter needs an explicit domain-gating fix before merging.

• The template can send a workspace API key to an external MCP host when the allowlist env var is unset or empty.
• The form-state and header-shape changes appear compatible with the existing save and test flow.
• The docs mention the allowlist only when it is already configured, which leaves the default permissive path unchanged.

apps/sim/app/workspace/[workspaceId]/settings/components/mcp/components/mcp-server-form-modal/mcp-server-templates.ts

Security Review

The new Unstructured starter pre-fills an external MCP URL with a bearer-token header sourced from a workspace env var. In deployments where ALLOWED_MCP_DOMAINS is unset or empty, the existing MCP domain policy allows that outbound secret-bearing traffic without an explicit allowlist entry.

Important Files Changed

Filename	Overview
apps/sim/app/workspace/[workspaceId]/settings/components/mcp/components/mcp-server-form-modal/mcp-server-templates.ts	Adds the Unstructured Transform starter template, including an external URL and bearer-token header default.
apps/sim/app/workspace/[workspaceId]/settings/components/mcp/components/mcp-server-form-modal/mcp-server-form-modal.tsx	Adds the starter template picker and applies selected template values through the existing MCP form state.
apps/docs/content/docs/en/agents/mcp.mdx	Documents starter configurations and the Unstructured Transform setup requirements.

_{Reviews (1): Last reviewed commit: "Clarify Unstructured Transform MCP templ..." | Re-trigger Greptile}

[greptile-apps]

Default-Allow External MCP Egress

When ALLOWED_MCP_DOMAINS is unset or empty, the existing MCP validation treats all domains as allowed. This new starter makes mcp.transform.unstructured.io a one-click external endpoint with Authorization: Bearer {{UNSTRUCTURED_API_KEY}}, so self-hosted deployments can send a workspace secret to that host without any explicit domain allowlist entry.

Rule Used: API keys and other user-provided credentials shoul... (source)

Learned From
simstudioai/sim#2133

[SudSampath]

@waleedlatif1, would you mind taking a look when you have a chance? This adds Unstructured Transform as a starter template in Sim's existing MCP setup flow.

I also opened the companion Unstructured docs PR at https://github.com/Unstructured-IO/docs/pull/995 so users have a first-party setup and workflow guide. The docs PR will stay draft until this lands and we can validate the deployed flow end to end.