Ai Agents Mcp
AI Agents / MCP
CleverOps can expose selected organization data and actions to external AI clients (Claude, Cursor, Windsurf, ChatGPT, and agent runners) via MCP (Model Context Protocol). This page explains what you can do, how access works, and how to connect a client.
Where you set it up
- In the app, go to Settings → AI Agents / MCP.
- Only owners/admins can create or revoke MCP keys.
- You’ll see the endpoint URL, a place to create a key, and a list of existing keys (with last-used timestamps).

How authentication works
- CleverOps uses a dedicated MCP API key. Your client sends it as an Authorization bearer token.
- The secret is shown once when you create it. Store it safely.
- If you revoke a key, it stops working immediately.
Header
Authorization: Bearer clevero_mcp_…
Endpoint URL
Use the endpoint shown in Settings → AI Agents / MCP. In staging it is typically:
https://mcp.stage.clevero.tech/mcp
It should end with /mcp.
Client configuration examples
The Settings page provides copy/paste snippets. A common format is a JSON config pointing to the remote HTTP MCP server:
{
"mcpServers": {
"clevero": {
"serverUrl": "https://mcp.stage.clevero.tech/mcp",
"headers": {
"Authorization": "Bearer <YOUR_MCP_API_KEY>"
}
}
}
}Recommended: store the key in an environment variable instead of committing it into a config file.
What the MCP server can do
MCP exposes tools that let external clients read context and perform actions inside your CleverOps workspace. Availability depends on what your workspace has configured and your membership.
Ask AI about anything in your workspace
- Ask questions and get answers using the same context and formatting rules as in-app AI.
- Useful for: “What’s our renewals risk?”, “Which tools are missing owners?”, “Summarize recent intelligence about Tool X”.
- AI usage is rate-limited the same way as in the app.
Inventory: tools
- List tools in your inventory (including basic vendor/status/renewal fields).
- Create a tool, update a tool, or archive a tool (where enabled).
- Enrich and add a tool from a URL (where enabled).
- Get aggregated tool stats (seats/cost/assignment counts) when available.
Org structure: departments and teams
- List departments and teams.
- Create or update departments/teams (where enabled).
- Pull department/team “summary” and “overview” rows used by the app screens (where enabled).
Signals: issues, events, and intelligence
- Create and update Issues (title, status, severity, links, deadlines, and metadata).
- List Issues and Intelligence items/events.
- Mark intelligence and events as read or dismissed for triage workflows.
- Fetch Events lists, details, and update event state (where enabled).
Portal: announcements, quick links, and sections
- Create/update/delete portal announcements (publishing requires an explicit confirmation flag).
- Create/update/delete portal quick links.
- Create/update/delete portal sections and reorder sections.
- Pull a full portal snapshot (settings + content lists) to audit what’s currently published.
Content studio: Plans, SOPs, Trainings, Reports
- Create Plans/SOPs/Trainings/Reports (optionally kick off generation runs).
- List and fetch items (including blocks + scope targets) for automation and audits.
- Update items (name/description/status/prompt/scope targets) and manage lifecycle (publish/archive).
- Edit content by creating/updating/deleting/reordering blocks.
Publishing to the Portal Library
- Publish a Report/Plan/SOP/Training to the portal library (internal or restricted visibility).
- For restricted visibility, you must provide at least one pack id.
- Unpublish items from the portal library.
Analytics and governance summaries
- Pull analytics overviews (governance, spend, utilization, renewals, connectors coverage/health) where enabled.
- Useful for building automated weekly summaries in external tools.
Recommendations workflows
- Start recommendation generation and check generation status (where enabled).
- List recommended replacements/new tools (where enabled).
- Support candidate review workflows: start review, add notes, approve, or reject (where enabled).
Training compliance and assignments
- Pull compliance summaries and drill into training/user detail (where enabled).
- List assignments for you (or a training) and mark assignments completed / attest (where enabled).

Operational notes
- Treat MCP keys like passwords. If a key is exposed, revoke it and create a new one.
- If an external client can’t connect, confirm the endpoint URL ends with /mcp and that the Authorization header is set.
- If you change membership (remove a user from the org), keys tied to that user will stop working.