Seam MCP

The Seam MCP server connects your AI assistant to Seam documentation, device information, API specifications, and task-specific implementation and troubleshooting guides. Sign in to also let your assistant read or manage live workspace resources through the Seam API, within the permissions you grant.

Choose Your Connection

Both endpoints use Streamable HTTP. Use the authenticated endpoint by default with an OAuth-capable client. It provides all discovery tools and prompts you to sign in for workspace access. The anonymous endpoint remains an option for documentation and skill packs without an account.

Use case	Server URL	Authentication
Read or manage workspace resources, plus all discovery tools	`https://mcp.seam.co/mcp/authenticated`	OAuth sign-in and Seam permission consent
Search public documentation, device information, API specifications, and guides	`https://mcp.seam.co/mcp`	None required for discovery

Connecting anonymously to /mcp does not automatically prompt you to sign in. Use /mcp/authenticated for workspace access. You do not need to put a Seam API key in your client configuration for OAuth setup.

One-Click Install in Cursor

Setup

The examples below use the authenticated endpoint. For public discovery without sign-in, replace the URL with https://mcp.seam.co/mcp and skip the login step. If you already have a Seam MCP entry, follow Update an Existing Installation first.

Run these commands in your terminal:

claude mcp add --transport http seam https://mcp.seam.co/mcp/authenticated
claude mcp login seam

You can also run /mcp inside Claude Code, select seam, and authenticate there. Reconnect the server or start a new session after signing in. See Claude Code’s MCP instructions.

Add to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all projects:

{
  "mcpServers": {
    "seam": {
      "url": "https://mcp.seam.co/mcp/authenticated"
    }
  }
}

Reload Cursor, open its MCP settings, and authenticate the seam connection. Start a new chat after signing in. See Cursor’s MCP instructions.

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "seam": {
      "serverUrl": "https://mcp.seam.co/mcp/authenticated"
    }
  }
}

Reload the MCP connection and complete the client’s OAuth sign-in. See Windsurf’s MCP instructions.

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "seam": {
      "type": "http",
      "url": "https://mcp.seam.co/mcp/authenticated"
    }
  }
}

Start the server from VS Code’s MCP controls and complete the OAuth sign-in when prompted. See VS Code’s MCP instructions.

Run these commands in your terminal:

codex mcp add seam --url https://mcp.seam.co/mcp/authenticated
codex mcp login seam

Alternatively, add to ~/.codex/config.toml, then run codex mcp login seam:

[mcp_servers.seam]
url = "https://mcp.seam.co/mcp/authenticated"

Restart your Codex session after signing in. Use /mcp to check the available tools. See OpenAI’s official MCP instructions.

Open ChatGPT Plugins, select the plus button, then Add custom MCP server. Name it Seam and enter https://mcp.seam.co/mcp/authenticated. Choose OAuth, complete Seam sign-in, and select Create as a plugin. Review the discovered tools and install the plugin. In a new conversation, type @ and select Seam.See OpenAI’s official MCP setup instructions for your ChatGPT surface.

Open Claude Desktop
Go to Customize → Connectors
Select + Add → Add custom connector
Enter: https://mcp.seam.co/mcp/authenticated
Connect the server, complete the browser sign-in, and enable it in a new conversation.

After adding the server, complete your client’s OAuth sign-in and refresh the connection so it loads the authenticated tool catalog.

Authorize Workspace Access

Sign in to Seam through your MCP client’s browser flow. Use the Seam account that has access to the workspace you want to inspect.
Ask for a read-only operation, such as listing your accessible workspaces. If Seam returns a permission-consent link, open it, choose the scope and permissions, approve, and retry the request.
Choose Viewer for read-only use. The Viewer permission bundle grants product.read, suitable for support triage and resource inspection. Grant write or management permissions only when you want your assistant to perform those operations.
Select the intended workspace. When more than one workspace is available, tell your assistant which one to use. Workspace-scoped requests can pass workspace_id explicitly.

Your assistant’s delegated access is limited by both the permissions and scope you approve and your own Seam access. Signing in identifies you; permission consent authorizes API execution. If a request needs an additional permission, review the grant in the Seam Console and retry after updating it.

Available Tools

The standard public server exposes seven read-only discovery tools. An authenticated connection also exposes api_request. A connection configured for a specific agent can expose a smaller tool set; use your client’s current tool listing as the source of truth.

Tool	Inputs	What it does
`search_docs`	`query`; optional `limit`, `min_score`	Searches Seam documentation and device information semantically. Returns ranked snippets and paths to read with `get_docs`.
`get_docs`	`globs`	Reads full documentation pages matching path globs. Use paths from `search_docs`; narrow patterns that match more than 20 pages.
`search_api`	`query`; optional `limit`	Finds documented API endpoints, resources, and related specification entries. Returns stable IDs; use `query: "*"` to browse the API inventory.
`get_apis`	`ids`	Reads detailed API specifications for IDs returned by `search_api`, including endpoint parameters and response shapes.
`search_prompts`	Optional `match`	Discovers implementation and troubleshooting guides by exact-match metadata, such as `match: {"task": "setup"}` or `match: {"task": "troubleshooting", "output": "direct"}`.
`get_prompt`	`name`	Loads a guide returned by `search_prompts`, its instructions, and supporting-file paths. Agent-specific connections can omit `name` to load their entrypoint.
`get_prompt_context`	`name`, `globs`	Reads a guide’s supporting files. Use the name and paths from `get_prompt`, following any instructions to load narrower paths. `globs: ["*/"]` loads all supporting files.
`api_request`	`endpoint`; optional `parameters`, `workspace_id`, `pagination`	Executes a documented Seam API endpoint selected through `get_apis`. Requires authentication and permission for the requested operation. Returns `error`, `resource`, and `pagination`.

search_docs and ordinary search_api searches default to five results and support a maximum of 20. With query: "*", search_api returns the full API inventory. search_docs also accepts a minimum confidence score through min_score. For paginated API requests, pass the returned pagination.next_page_cursor as pagination.page_cursor on the next call.

api_request supports both reads and writes. A Viewer grant enforces read-only API access; the tool itself is not limited to retrieval. The other seven tools read public information and do not access your private workspace resources.

How the Tools Work Together

Answer a documentation question: search_docs → get_docs. Read the full page before relying on a search snippet. For device support, check the returned integration_support_level; a device appearing in search is not proof that Seam integrates it.
Inspect or manage workspace resources: search_api → get_apis → authenticated api_request. Your assistant discovers the endpoint, reads its specification, and executes it within your grant. Workspaces, devices, access codes, thermostat schedules and presets, and connected accounts are reached through the API tool rather than a separate MCP tool for every resource.
Follow an implementation or troubleshooting guide: search_prompts → get_prompt → get_prompt_context. Load the guide and the supporting context it requests before following its steps. For a live incident, the troubleshooting guide then uses the authenticated API tools to collect evidence.

Troubleshoot with a Skill Pack

The public MCP includes a troubleshooting skill pack that guides your assistant through investigating a concrete incident. It inspects public API evidence, explains what the evidence establishes, and recommends a next step. When public evidence cannot establish the cause, it identifies the gap and prepares a summary you can share with Seam support. You can load the instructions anonymously. To inspect your workspace, use the authenticated connection and grant Viewer access. If you already use that connection, reconnect it and start a new conversation to load the latest server guidance.

What It Covers

Playbook	Example symptoms
Device and connected-account health	A device is offline, missing, or stale; a connected account needs reauthorization.
Access grants and provisioning	Access did not arrive, reached only some doors, or has unexpected timing.
Access codes and timing	A code is stuck setting, failed to set, or reportedly still works after its end time.
API errors and MCP authentication	An API request failed, `api_request` is missing, or login or permission consent is blocked.

The assistant loads only the relevant playbook and follows related resources when evidence calls for it. For access created through an Access Grant, it starts with the grant before inspecting a related low-level access code.

Try an Investigation

Give your assistant the workspace ID, the affected resource ID when available, and what happened. Include the incident time and timezone if known, especially for timing problems. Replace the placeholders in these prompts with your own IDs:

“Use Seam MCP to troubleshoot why device <device_id> in workspace <workspace_id> went offline at <time with timezone>.”
“Access grant <access_grant_id> in workspace <workspace_id> should have worked at <time with timezone>, but the guest could not enter. Investigate what happened.”
“Access code <access_code_id> in workspace <workspace_id> is stuck setting. What does the public evidence show, and what should I do next?”
“My Seam MCP connection uses https://mcp.seam.co/mcp. I see documentation tools but no api_request, and was never prompted to sign in. Why?”

Connection failures can be investigated before workspace access is available. Resource investigations require workspace scope; if the assistant cannot identify one affected resource unambiguously, it asks for the smallest missing detail.

The troubleshooting pack instructs your assistant to keep all operations read-only. It recommends fixes without changing devices, codes, grants, or account connections. Choose Viewer to enforce read-only API access through your delegated permissions. The pack’s instructions do not restrict the general-purpose api_request tool for other tasks.

Load the Skill Explicitly

The MCP server guides your assistant toward troubleshooting for incident questions. Your assistant chooses which tools to call; a phrase alone does not guarantee that it loads the pack. To request it explicitly, ask:

Call Seam search_prompts with task: troubleshooting and output: direct, then load the returned guide and its relevant playbook before investigating.

The tool arguments for search_prompts are:

{
  "match": {
    "task": "troubleshooting",
    "output": "direct"
  }
}

Next, call get_prompt with name: "tech_support_agent-direct". Follow the returned instructions to load the playbook index, evidence model, and relevant supporting files through get_prompt_context. The guide uses search_api → get_apis → api_request for live workspace inspection. For a dedicated troubleshooting connection, use https://mcp.seam.co/mcp/authenticated?agent=tech_support. On that connection, get_prompt can be called without a name to load the troubleshooting entrypoint. For an integration that needs a structured result, search with output: "json" and load tech_support_agent-orchestrated. Its instructions include a JSON output schema. The dedicated JSON entrypoint is https://mcp.seam.co/mcp/authenticated?agent=tech_support_json.

Understand the Result

The investigation distinguishes observed facts from an established cause. Current state does not by itself prove what happened earlier, and an empty event history does not prove an event never occurred. If the cause remains unresolved, the assistant summarizes the inspected resources, relevant timestamps, sanitized errors, and missing evidence for a support handoff. It does not submit a ticket or contact support for you. Keep API keys, tokens, PIN values, and instant-key links out of prompts and shared findings.

Try It Out

For public discovery:

“What access control systems does Seam support? Read the relevant documentation.”
“Find the API specification for listing thermostat schedules.”
“Find the project setup guide for Seam.”

After signing in and granting Viewer access:

“Use the Seam MCP to list my accessible workspaces. Keep all operations read-only.”
“In this workspace, list the connected accounts and devices with errors or warnings.”
“Read this thermostat’s configured climate presets and schedules. Do not change anything.”
“Show the access codes for this device without creating or updating any codes.”

Update an Existing Installation

If you installed the original local Seam MCP, replace its local command or stdio configuration with the hosted Streamable HTTP URL. Reinstalling the old local package does not migrate it to the hosted server. If your existing hosted entry uses /mcp, change it to /mcp/authenticated when you need workspace access. For Claude Code, remove the old entry using its actual configured name and add the hosted connection:

claude mcp remove seam
claude mcp add --transport http seam https://mcp.seam.co/mcp/authenticated
claude mcp login seam

For Codex:

codex mcp remove seam
codex mcp add seam --url https://mcp.seam.co/mcp/authenticated
codex mcp login seam

For Cursor and other editors, replace the existing Seam entry with the corresponding setup configuration. Check both project and user configuration for old entries. Reconnect or restart the client after authentication, then start a new conversation so the assistant receives the current tools.

Troubleshooting

No sign-in prompt, or only seven tools? Verify that the configured URL ends in /mcp/authenticated, complete your client’s OAuth login, and refresh the connection. An anonymous /mcp connection successfully loads discovery tools without requesting sign-in.
Signed in, but API access is refused? Complete the Seam permission-consent link returned on first API use and retry. For a missing-permission error, review the grant’s permissions and scope in the Seam Console.
api_request is still missing? Reconnect or restart your client after login to refresh its tool catalog. Check for a stale local server entry or an agent-specific connection with a restricted tool set.
The assistant expects get_doc or list_doc_sections? Those are legacy tools. Refresh the connection and use search_docs → get_docs.
Several workspaces are available? Select the intended workspace and pass its workspace_id for scoped requests.
Browser sign-in does not open on an SSH or headless host? Current Claude Code and Codex clients support mcp login seam --no-browser. Follow the client’s instructions to open the authorization URL in a browser and complete the callback.
Connection errors? Use an HTTPS URL with the full /mcp or /mcp/authenticated path and the client’s HTTP transport. Check that your network allows outbound HTTPS to mcp.seam.co.

For help, contact support@seam.co.

Getting Started

Use Cases

Core Concepts

Capability Guides

UI Components

Developer Tools

Industry Guides

Device Manufacturer Guidance

Low-Level APIs

Choose Your Connection

One-Click Install in Cursor

Setup

Authorize Workspace Access

Available Tools

How the Tools Work Together

Troubleshoot with a Skill Pack

What It Covers

Try an Investigation

Load the Skill Explicitly

Understand the Result

Try It Out

Update an Existing Installation

Troubleshooting

​Choose Your Connection

​One-Click Install in Cursor

​Setup

​Authorize Workspace Access

​Available Tools

​How the Tools Work Together

​Troubleshoot with a Skill Pack

​What It Covers

​Try an Investigation

​Load the Skill Explicitly

​Understand the Result

​Try It Out

​Update an Existing Installation

​Troubleshooting

Choose Your Connection

One-Click Install in Cursor

Setup

Authorize Workspace Access

Available Tools

How the Tools Work Together

Troubleshoot with a Skill Pack

What It Covers

Try an Investigation

Load the Skill Explicitly

Understand the Result

Try It Out

Update an Existing Installation

Troubleshooting