Install the Speak AI plugin in any AI coding client

Install the Speak AI plugin in Claude Code or any Agent Plugins compatible client, with the marketplace commands, the portable path, and OAuth consent.

You need a Speak AI account. Some paths also need an API key, which you create at app.speakai.co/developers/apikeys.

What does the Speak AI plugin install?

Installing the plugin gives your agent the 112 tools, 5 resources, and 3 prompts the Speak AI MCP server exposes, plus five skills that tell the agent how to use them: getting-started, meeting-summaries, research-analysis, clips-and-captions, and automations-and-webhooks.

The plugin connects to Speak AI over the internet, so nothing runs on your machine and no API key is stored on disk.

Why do install steps differ between clients?

Each client decides how a plugin reaches your machine and how you turn it on. One gives you a marketplace command, another asks you to point at a Git repository, and a third asks you to copy a directory into a plugins folder.

What you get is the same in every case, so follow whichever set of steps matches your client below.

How do you install the plugin in Claude Code?

Claude Code installs the Speak AI plugin with two commands, one to add the marketplace and one to install the plugin from it.

Add the marketplace and install the plugin:

claude plugin marketplace add speakai/speakai-mcp
claude plugin install speakai-mcp@speakai

When Claude Code enables the plugin, enter your Speak AI API key from:
```
https://app.speakai.co/developers/apikeys
```
Confirm it worked. Run /mcp inside Claude Code and check that the speakai server shows as connected.

Claude Code also lists Speak AI in the official plugin marketplace under a different name and install command. See Connect Claude Code to Speak AI for that route.

How do you install the plugin in any other Agent Plugins client?

Any Agent Plugins compatible client installs the Speak AI plugin from the plugins/speakai-mcp directory of the speakai/speakai-mcp repository on GitHub. Follow your client’s own instructions for adding a plugin, and give it that repository and path.

The directory your client reads looks like this:

plugins/speakai-mcp/
  plugin.json
  mcp.json
  skills/
    getting-started/SKILL.md
    meeting-summaries/SKILL.md
    research-analysis/SKILL.md
    clips-and-captions/SKILL.md
    automations-and-webhooks/SKILL.md

Once the client has the directory, it reads mcp.json, connects to https://api.speakai.co/v1/mcp over streamable HTTP, and starts the OAuth consent step described below.

The OAuth consent step is where you sign in to Speak AI in a browser and approve the client’s request, and the client stores the resulting token so you never paste an API key. The Speak AI MCP server supports OAuth 2.1 with Dynamic Client Registration, which is what lets a client you have never used before register itself and connect in one click.

What you see is short:

Your client opens a browser window pointed at Speak AI.
You sign in, or click Confirm if you are already signed in.
You approve the request, and the browser hands the token back to your client.

Approving grants the agent read and write access to your whole Speak AI workspace, which includes viewing, modifying, and deleting recordings, folders, and clips. Access is not continuous. The agent reads data only when it calls a tool to answer something you asked, and your recordings stay in your Speak AI workspace.

To disconnect, remove the plugin in your client, or revoke the connection at api.speakai.co/v1/oauth/connections.

The same endpoint accepts a Speak AI API key as a Bearer token, so a client with no browser can still connect. Send the key on every request:

Authorization: Bearer speak_sk_example_000000000000

For stdio mode and the command line, set the key as an environment variable instead:

export SPEAK_API_KEY="speak_sk_example_000000000000"

Create the key at app.speakai.co/developers/apikeys. See MCP authentication for token handling, rate limits, and the tool error format.

How do you confirm the plugin is working?

Ask your agent for something small that needs a Speak AI tool, and check that it answers from your workspace. A request like “list my five most recent recordings” makes the agent call list_media, which fails loudly if the connection or the credentials are wrong.

Good first requests:

List my five most recent recordings.

Summarize this week's team meetings into decisions, action items, owners, and unresolved risks.

Find a strong 30-second highlight from the latest webinar, create a clip, and export captions.

You do not need to name tools. The agent picks them from what you ask.

What do you do when tools do not appear?

Tools missing after a successful install almost always mean the client has not reloaded the plugin yet. Restart the client, or run its reload command, then check the plugin list again.

Three failures cover most cases:

No Speak AI tools listed. The plugin is installed but not activated. Reload plugins, then check your client’s MCP server list for speakai.
Requests fail with an authorization error. The token or key is no longer valid. Revoke the connection at api.speakai.co/v1/oauth/connections and connect again, or create a new API key.
The connection is blocked by policy. Some Claude Team and Enterprise workspaces restrict custom connectors. Your workspace admin needs to allow https://api.speakai.co/v1/mcp.