The Speak AI Agent Plugin is a single package that connects an AI agent to a Speak AI workspace and teaches it how to work there. It connects your agent to the Speak AI MCP server and adds 5 skills. Install it once and your agent gets all 112 tools plus written instructions for using them.
What is the Speak AI Agent Plugin?
The Speak AI Agent Plugin packages the Speak AI MCP server connection together with the skills an agent needs to use it well. The MCP server exposes 112 tools, 5 resources, and 3 prompts across 15 categories covering media, transcripts, AI insights, folders, recorders, clips, exports, automations, webhooks, dashboards, and team management. The plugin adds the missing half: written procedures that tell an agent which tools to call, in what order, and what to do when something is still processing.
The plugin is named speakai-mcp and its current version is 1.18.0, the same version as the @speakai/mcp-server package that implements the server.
Does the plugin work in more than one agent?
Yes. It follows the open Agent Plugins standard, so the same plugin works in every agent that supports the format rather than being tied to one. You install it once and it keeps working as you move between tools.
What does a plugin give you that a manual MCP setup does not?
A manual MCP setup gives an agent access to 112 tools, and the plugin gives it access plus the knowledge of how to use them. That is the entire difference, and it is larger than it sounds.
When you paste the server URL into a client by hand, the agent sees 112 tool names and their parameter schemas. It does not know that a recording has to finish processing before you can read its transcript, that a research question spanning many recordings should be scoped to a folder before you ask it, or that cutting a clip means finding the moment in a timestamped transcript first. It guesses. Sometimes it guesses well, and sometimes it calls the wrong tool three times and reports that the data is missing.
The skills close that gap. Each one is a written procedure for a real job, listing the exact tools to call in order, the failure cases to expect, and what to do about each. The agent reads the skill when the request matches it, then calls tools with a plan instead of a guess.
You also get versioning and updates. The plugin is a versioned package, so a new release ships new tools and updated procedures together. A hand-configured server URL never tells you that anything changed.
If you only want raw tool access and you are writing the orchestration yourself, configuring the MCP server by hand is still a reasonable choice. See MCP server setup for that path, or the Node SDK if you want to call the tools from your own code.
Which skills come with the plugin?
The plugin ships 5 skills, each covering one kind of work an agent does inside a Speak AI workspace.
| Skill | What it covers |
|---|---|
getting-started |
Connecting the agent, orienting it in the workspace, and finding the right tool among the 15 categories. |
research-analysis |
Finding themes, quotes, and sentiment across many interviews or customer calls, with citations. |
meeting-summaries |
Sending the assistant to a Zoom, Google Meet, or Microsoft Teams call, then producing decisions, action items, owners, and risks. |
clips-and-captions |
Cutting highlight clips, exporting captions and transcripts, and publishing embeds. |
automations-and-webhooks |
Building automations that run on their own, and moving data in and out with inbound and outbound webhooks. |
Every skill names real tools only, such as search_media, get_transcript, ask_ai_chat, create_clip, and schedule_meeting_event.
How do you install the plugin?
You install the plugin in two commands: add the marketplace that hosts it, then install the plugin from that marketplace. In Claude Code:
claude plugin marketplace add speakai/speakai-mcp
claude plugin install speakai-mcp@speakaiInside a running Claude Code session, use the slash command form instead:
/plugin marketplace add speakai/speakai-mcp
/plugin install speakai-mcp@speakai
/reload-pluginsAfter the plugin loads, ask for something and let the agent pick its own tools:
Find the last 10 customer interviews that mention pricing, group the feedback
by theme, and cite the source recordings.The full command set for each client, including the local development flow and what to check when tools do not appear, is on install the plugin.
Do you need an API key to use the plugin?
It depends which route you install through, and the two differ.
On the portable Agent Plugins route, you do not need a key. mcp.json points at the remote server, which authenticates with OAuth 2.1 and Dynamic Client Registration. Your client registers itself, a permission popup opens, you approve it once, and no key is ever pasted or stored by you.
On the Claude Code marketplace route, you do need one. Claude Code reads the native .mcp.json, which runs the server locally over stdio, so it prompts for your key on install. Generate one at app.speakai.co/developers/apikeys.
Three ways to authenticate exist and they are not interchangeable:
- The remote MCP endpoint
https://api.speakai.co/v1/mcptakes OAuth, orAuthorization: Bearer <speak-api-key>. - stdio mode and the CLI take the
SPEAK_API_KEYenvironment variable. There is no header to set. - The REST API takes two headers,
x-speakai-keyandx-access-token. It does not accept a Bearer token. See authentication.
See Authentication for the token flow, rate limits, and the tool error format.
Where do you look up a specific tool?
Every one of the 112 tools has its own reference page under the tool reference, grouped by the same 15 categories the skills use. Per-tool pages follow the pattern https://docs.speakai.co/mcp/tools/<category-id>/<tool_name>/, so get_transcript is at /mcp/tools/media/get_transcript/.
You do not need to memorize tool names to use the plugin. The skills handle tool selection, and the reference exists for when you want to check a parameter or a return shape yourself.