Build with Pi. Build a device integration in your project.

Add Seam’s current documentation to your Pi session and build a device integration. Start with a small script, verify the API behavior, and move the working code into your application.

Set up Pi Troubleshoot an issue

Pi

Illustrative sandbox workflow

Set up Seam in Pi.

Current Pi releases include MCP support. Add Seam’s hosted endpoint, then check it before starting the application setup.

01

Register the Seam endpoint

With a current Pi installation, run this in your terminal. Complete browser sign-in. Pi saves servers at user scope by default; add --local for a trusted project’s .pi/mcp.json.

Run in your terminal

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

02

Check discovery

List the connections and confirm seam exposes its documentation tools. Inside an existing session, run /reload after adding the server; /mcp shows connection details.

Verify the connection

pi mcp list

03

Start with the current setup guide

Paste this into Pi from your application directory. Ask it to read /agents.md and use the current documentation to plan your integration.

Seam agent setup

Paste into Pi

Set up Seam in this project:
fetch https://seam.co/agents.md and follow its instructions.

Choose your workspace permissions.

After signing in, check that api_request appears alongside the documentation tools. Ask your agent to list your workspaces. If Seam returns a permission-consent link, open it, select the intended scope and Viewer for read-only access, then retry. Device changes require write permissions. No Seam API key belongs in this MCP configuration.

Already connected? Replace the old Seam entry instead of adding a duplicate, reconnect or restart your client, and start a new conversation to refresh its tools. For documentation and skill packs without sign-in, use https://mcp.seam.co/mcp instead and skip OAuth login. This optional anonymous connection cannot access workspace resources.

Read the MCP guide

Troubleshoot with your agent.

Investigate device and connected-account issues, Access Grants, access codes, and API or authentication errors. Seam’s MCP delivers the troubleshooting skill pack directly to your agent, with playbooks for gathering evidence and identifying what is still unknown.

Connect with Viewer access.

Use an MCP client that supports OAuth to connect to Seam’s authenticated endpoint. Sign in and approve Viewer access to the workspace you want to investigate. Refresh the connection and start a new conversation to load its tools.

Use your existing workspace. This skill is available through MCP without a separate plugin install. Replace the placeholders with the affected resource, workspace, and incident time. The investigation uses public API evidence and stays read-only.

Set up troubleshooting

Authenticated MCP endpoint

https://mcp.seam.co/mcp/authenticated

Try this prompt

Use Seam MCP to investigate why access grant <grant_id>
in workspace <workspace_id> failed at <time and timezone>.
Use the troubleshooting skill pack. Keep the investigation
read-only and explain any missing evidence.

Prompts to try.

Give Pi a concrete task, ask it to check current documentation, and review the result.

Build a small working script

Agent prompt

Use Seam’s current documentation to create the smallest device-listing script that fits this repository. Read the Seam API key from the environment, run it against my sandbox, and explain the returned identifiers before adding any write operation.

Turn a script into an application service

Agent prompt

Take our working Seam sandbox script and move it into the application’s service layer. Preserve current API types, surface actionable errors, and keep the environment setup outside the source code.

Verify one complete access lifecycle

Agent prompt

Use the current Seam Access Grants documentation to implement a sandbox lifecycle check: create a grant, inspect its readiness, change its time window, revoke it, and clean up. Report the observed result of each step.

Questions.

Client setup references

Pi MCP documentation ↗Pi skills ↗Seam agent setup ↗Seam integration skills ↗

What if my Pi install does not recognize pi mcp?

Check your installed version against the current Pi documentation. Current releases include built-in MCP support. An extension that registers /mcp can replace that support inside sessions, so check extensions if shell and session behavior differ.

Can Pi load the Seam skills?

Yes. Pi supports portable SKILL.md directories, including project .agents/skills/. Keep each Seam skill’s supporting files with it. Use /skill:seam-access-grants to explicitly load that installed skill, and /reload after adding skills during a session.

Where does Pi save this connection?

User connections live in ~/.pi/agent/mcp.json. Project connections live in .pi/mcp.json and load after project trust is granted. A project entry with the same name can override the user entry.

Next steps.

Test in a sandbox, then connect your real devices.

Find your device

Get an API key All agent guides

Build with Pi. Build a device integration in your project.

Set up Seam in Pi.

Register the Seam endpoint

Check discovery

Start with the current setup guide

Troubleshoot with your agent.

Connect with Viewer access.

Prompts to try.

Build a small working script

Turn a script into an application service

Verify one complete access lifecycle

Questions.

What if my Pi install does not recognize pi mcp?

Can Pi load the Seam skills?

Where does Pi save this connection?

Next steps.

Global Infrastructure for Controlling the Physical World