MCP servers
Add your company's own remote MCP server, review each of its tools with a risk tier, and let agents use the ones you approve.
If your company runs a remote MCP (Model Context Protocol) server, Everdeck can give its tools to your agents. Agents use only the tools a company admin approves, each with a risk tier, under the same approval rules as every other tool. MCP servers count toward your plan's MCP and API servers. Company admins manage them on the MCP servers page.
Add a server
Select Add MCP server and fill in:
- Name: what people see, like "Customer records".
- Key: a short name agents see its tools under:
mcp_<key>_<tool>. It can't be changed later. - Address: the server's https address. Everdeck keeps the full address sealed and shows only its host and path, since some addresses hold a secret.
- Transport: Streamable HTTP or SSE (legacy), or let Everdeck choose from the address.
- Authentication: None, A header (a Header name and Header value), A bearer token (a Token), or Your OAuth app.
- Call timeout (seconds): how long one tool call may take.
Saving asks you to confirm it's you, before you type any credential.
Sign in with your own OAuth app
For Your OAuth app, select Discover from the address to fill in the server's sign-in details, or enter them: the Client ID and Client secret of an OAuth app you registered with the server's authorization server, the Authorization address, the Token address, Scopes separated by spaces, the Client authentication and the Resource. Register this redirect URI with your OAuth app:
https://everdeck.io/api/mcp-servers/oauth/callback
Save the server, then select Sign in on its page.
Review its tools
Once Everdeck can reach the server, it reads the server's list of tools. Each one waits for review: open the server and look at its Tools. For each tool you see its name as agents see it, what the server says it does and the input it takes, any hints the server gives (for example that it's read-only or destructive), the tier Everdeck suggests, and the tier you choose.
- Choose its tier, decide whether its output is Untrusted output (on unless the tool returns only your company's own data), and select Approve, or Reject it.
- Approve all waiting approves every waiting tool at once, each at tier 3 or higher (its suggested tier when that's higher), with its untrusted setting as it is.
Agents get a tool only once it's approved, and only when their charter grants it on the agent's Tools tab. Agents always get the definition you approved, never the server's live one. If the server changes a tool you approved, it shows Changed with the approved and current versions side by side, and agents can't use it until you approve the new version.
Keep it healthy
Everdeck checks each server regularly, and Check now checks it at once. A server whose checks or calls fail several times in a row shows Needs attention, with the last error. Pause stops agents using it until you Resume. Edit changes it: a new address needs its credentials again, and its tools are approved again for the new address. Delete removes it and its tools.
Everything a server says about itself and its tools is shown as plain text. Its own instructions, if it has any, are shown for reference only: agents don't get them.
More in Connecting apps and tools
- Connecting appsConnect the apps your agents work in, with a key or token or with your own OAuth app, and test, fix, disconnect and revoke connections.
- Intake: webhooks, watched endpoints and rulesTurn requests from other apps into tasks for your agents, with webhooks, watched API endpoints, connector triggers and rules.
- Jira syncBring Jira issues to your agents as tasks, and have Everdeck write status and finished results back to Jira.
- API servers (OpenAPI)Import your own API from its OpenAPI document, choose which operations agents may use, and set each one's risk tier.
Still stuck?
Email [email protected] and say what you were trying to do. Or browse every guide.