Gemini + Specky
Use Specky with the Gemini API and Remote MCP
Supported Gemini API and agent runtimes can call Specky’s Streamable HTTP MCP endpoint as an mcp_server tool. Configure it server-side, use the lowercase name specky, and restrict allowed_tools to the smallest useful set.
Canonical MCP endpoint
https://www.specky.space/api/mcpStreamable HTTP · OAuth 2.1 with PKCE · workspace key fallback
Setup
Connect Gemini to Specky
Follow the provider-specific path below. The exact menu names and availability can change as each provider evolves its MCP and directory programs.
- 01
Configure a remote MCP tool
Create a Gemini interaction or agent with an mcp_server tool whose URL is https://www.specky.space/api/mcp. Use the server-side API or managed-agent environment; do not put workspace credentials in browser code.
- 02
Use a valid server name
Set the MCP server name to specky. Keep it lowercase and avoid a hyphen so it satisfies the current Gemini naming rules.
- 03
Allow only the tools you need
For a first run, allow search_workspace and read_workspace_item, or the public feedback_to_spec tool. Add write tools only after the read path is understood.
- 04
Verify the runtime’s MCP support
Google’s MCP support is evolving across Interactions, Agents, models, and preview environments. Check the runtime documentation and run an initialize/tools-list smoke test before relying on it in production.
tools: [{
type: "mcp_server",
name: "specky",
url: "https://www.specky.space/api/mcp",
allowed_tools: ["search_workspace", "read_workspace_item"]
}]Use provider-managed secrets or OAuth for authenticated workspaces. Do not paste a workspace key into public client-side code.
What to test first
Use one bounded, read-first prompt so you can inspect the returned evidence and confirm the workspace boundary:
Find the strongest evidence behind this customer problem, show the sources and unknowns, and draft a problem statement. Do not create tickets or change records until I approve the draft.
Questions about Specky and Gemini
Does every Gemini model support remote MCP?+
No. Remote MCP support depends on the Gemini API or agent runtime and the model or preview surface in use. Confirm the current Google documentation for the runtime you are deploying.
How should Gemini authenticate to Specky?+
Use server-side credentials or the runtime’s managed credential mechanism. Never expose a Specky workspace key in client-side JavaScript or a public prompt.
Why should I restrict allowed_tools?+
An explicit allowlist reduces accidental access and makes the agent’s behavior easier to audit. Start read-only, then add a write action only when its approval flow is clear.