Everix MCP server: tools, keys and limits

Reference for connecting an AI assistant to Everix through the Model Context Protocol: Everix Cloud IFR and Cloud Risk, and installed Everix.

Last reviewed

Where it runs

Read-only MCP is available in every Everix option: Everix Cloud IFR and Everix Cloud Risk at /api/mcp, and installed Everix (On-site and Private Cloud) at /api/public/mcp. For what risk and finance teams ask it, see AI agents for risk and finance.

In one minute

  • Everix Cloud: https://app.everix.io/api/mcp, with an agent key from Organization › AI agents. IFR and Cloud Risk tools, filtered by the modules the key may read.
  • Try it: https://app.everix.io/api/mcp/demo needs no key and serves the IFR tools over the fictional Contoso Investments demo.
  • Installed Everix: https://<your-everix-host>/api/public/mcp, with a service-account key and “Allow MCP access” switched on.
  • Read-only and audited: 500 rows or 256 KB per result, 120 requests per minute per key, and every call recorded.

Everix Cloud IFR and Cloud Risk

The Everix Cloud portal serves the MCP server for your organization. One key reads one organization, and only the modules it was created for.

Connect a client

{
  "mcpServers": {
    "everix-cloud": {
      "type": "http",
      "url": "https://app.everix.io/api/mcp",
      "headers": { "Authorization": "Bearer ${EVERIX_CLOUD_KEY}" }
    }
  }
}

The exact format depends on the client. The AI agents page of your organization shows the endpoint and this snippet. Keep the key in an environment variable or the client's secret store, never in a shared file.

Agent keys

  1. An admin of your organization opens Organization › AI agents. Other roles do not see the page.
  2. They name the key, choose the modules it may read (IFR, Cloud Risk or both) and an expiry of 30, 90, 180 or 365 days. Up to 20 keys can be active at once.
  3. The key starts with evx_ and is shown once; Everix stores only a hash of it. Revoking a key takes effect on the next request. Creating and revoking keys is written to the organization's audit trail.

The scopes are read:ifr and read:cloud-risk; there are no write scopes. A key reads a module only while your organization uses it.

Tools

Every tool carries the MCP read-only annotation. A key sees only the tools of the modules it may read.

whoami
What it returns
Organization, key, scopes, readable modules and limits
Scope
any
ifr_list_entities
What it returns
Reporting entities with jurisdiction, IFR class and reference dates
Scope
read:ifr
ifr_get_latest_calculation
What it returns
The current calculation: own funds requirement and its binding component, own funds, headroom, ratio, class test, liquidity and finding counts
Scope
read:ifr
ifr_get_k_factors
What it returns
Each K-factor with its average, coefficient, value, window, legal reference and observations
Scope
read:ifr
ifr_list_findings
What it returns
Findings of a calculation: severity, code, message and reference
Scope
read:ifr
ifr_list_reports
What it returns
Report and export files of a calculation, with portal download links
Scope
read:ifr
ifr_list_runs
What it returns
Calculation history, newest first, with status
Scope
read:ifr
cloud_risk_get_inputs_status
What it returns
Whether your inputs are complete, and what is missing where
Scope
read:cloud-risk
cloud_risk_list_runs
What it returns
Risk runs, newest first
Scope
read:cloud-risk
cloud_risk_get_run
What it returns
One run: status, base currency, total PV, VaR, stress and priced trades
Scope
read:cloud-risk
cloud_risk_get_var
What it returns
VaR by method (Monte Carlo, historical), horizon and confidence
Scope
read:cloud-risk
cloud_risk_get_worst_scenarios
What it returns
The worst scenarios with total and per-trade P&L
Scope
read:cloud-risk

IFR tools pick a calculation by run, or by entity and period, and otherwise read your newest valid calculation. Download links point to the portal: a person signs in to download, and the key never downloads files. Every tool call is recorded, and the AI agents page lists the latest calls.

Public demo endpoint

https://app.everix.io/api/mcp/demo answers without a key, with the IFR tools over the same fictional firm as the guided IFR demo. It is read-only, holds synthetic data only and allows 30 requests per minute per address.

Installed Everix (On-site and Private Cloud)

On installed Everix the MCP server runs on your own host, next to the public REST API (/api/public/v1). Each tool maps one-to-one to a REST endpoint and uses the same permission checks.

Connect a client

{
  "mcpServers": {
    "everix": {
      "type": "http",
      "url": "https://<your-everix-host>/api/public/mcp",
      "headers": { "Authorization": "Bearer ${EVERIX_API_KEY}" }
    }
  }
}

Service accounts and keys

  1. An Everix administrator creates a service account in Admin › API access and chooses its scopes. An administrator can grant only scopes whose permissions they hold.
  2. They create an API key for it. The key starts with evx_ and is shown once. Expiry is 90 days by default and at most 365 days. Keys can be rotated with an overlap period and revoked at any time.
  3. They switch on Allow MCP access for that service account. It is off by default, so a key made for another integration cannot simply be pasted into an agent.

Tools

The tool list a client sees is filtered by the key's scopes: a key without read:audit does not see list_audit_events, and calling it anyway is refused.

All nine tools carry the MCP read-only annotation. REST paths are relative to /api/public/v1.

whoami
What it returns
The calling service account and its scopes
Scope
any
REST equivalent
GET /me
list_universes
What it returns
Calculation universes (result sets), filterable by date and type, with a cursor
Scope
read:results
REST equivalent
GET /universes
get_universe
What it returns
One universe and its metadata
Scope
read:results
REST equivalent
GET /universes/{id}
list_aggregations
What it returns
The aggregation views available on a universe
Scope
read:results
REST equivalent
GET /universes/{id}/aggregations
get_aggregation_tree
What it returns
An aggregation tree (for example XVA by counterparty and netting set), with depth, node and metric filters; capped
Scope
read:results
REST equivalent
GET /universes/{id}/aggregations/{agg}/tree
list_workflow_templates
What it returns
Configured calculation workflow templates
Scope
read:workflows
REST equivalent
GET /workflows/templates
list_workflows
What it returns
Workflow runs and their status
Scope
read:workflows
REST equivalent
GET /workflows
get_workflow
What it returns
One workflow run
Scope
read:workflows
REST equivalent
GET /workflows/{id}
list_audit_events
What it returns
Audit events by time range and type; capped
Scope
read:audit
REST equivalent
GET /audit/events

Every MCP tool call is written to the Everix audit trail as an MCP_TOOL_CALL event, with the tool name, a hash of the arguments, the result size and the outcome. A service account with read:audit can read these events back through list_audit_events or the REST API.

Protocol

  • MCP protocol revision 2026-07-28; clients that negotiate 2025-11-25 are served down-level.
  • Streamable HTTP, stateless: no session identifiers.
  • Tools only: no resources and no server-provided prompts.
  • Results are returned as structured content (JSON fields), never mixed into tool descriptions. Tool descriptions are static and versioned with the code.
  • Both servers publish Protected Resource Metadata (RFC 9728), and a 401 response carries a WWW-Authenticate header pointing to it.

Limits

Lists over 500 rows are cut with truncated: true and a cursor. A result over 256 KB is a tool error: narrow the query. Over a request limit, the server answers HTTP 429 with Retry-After. An administrator can change the installed defaults.

Rows per result
Everix Cloud
500, then truncated: true and a cursor
Installed Everix (default)
500, then truncated: true and a cursor
Result size
Everix Cloud
256 KB per tool result
Installed Everix (default)
256 KB per tool result
Requests per key
Everix Cloud
120 per minute
Installed Everix (default)
120 MCP tool calls per minute; 600 REST requests per minute
Failed authentication
Everix Cloud
20 per minute per address
Installed Everix (default)
20 per minute per client IP
Public demo
Everix Cloud
30 requests per minute per address
Installed Everix (default)
None

Errors

No key, invalid, expired or revoked key
Response
401, WWW-Authenticate: Bearer with resource_metadata
Installed: service account without “Allow MCP access”
Response
403 mcp-not-allowed
Tool outside the key's scopes or modules
Response
Refused; the tool is not in the key's tool list
Request from a foreign Origin
Response
403
Unknown object, or no data yet
Response
Tool error the agent can read, without internal details
Result over 256 KB
Response
Tool error asking to narrow the query (filters, limit, cursor or depth); not a 429 and not a truncated page
Rate limit exceeded
Response
429 with Retry-After

What the server does not do, on purpose

  • No write tools: uploads, calculations, approvals and settings stay with people.
  • No free-form query tool (no raw OData filters or database queries) and no bulk export.
  • No URL-fetch or file tools, and no admin tools: keys are managed in the portal or the admin screens only.
  • No cookies: the endpoints ignore the browser session and never use the server's own privileges on behalf of a caller.
  • No data from another organization: everything a Cloud key reads is built from its own organization on the server.

Recommended setup: one read-only key per agent, with the narrowest modules or scopes that answer its questions. For the reasoning behind these choices, read Giving AI agents safe access to risk data with MCP.

How to get access

  • Everix Cloud: an admin of your organization creates a key under Organization › AI agents and you connect your client. No request to Everix needed.
  • Installed Everix: book a 30-min call. We enable the API module and MCP on your instance; your administrator creates a read-only service account and key.

FAQ

Is the Everix MCP server available?+

Yes, read-only. Everix Cloud IFR and Everix Cloud Risk serve it at /api/mcp for every organization; an org admin creates the agent keys. On installed Everix (On-site and Private Cloud) it runs on your own host next to the public REST API.

Can I try it before we have an account?+

Yes. The public demo endpoint /api/mcp/demo needs no key and answers the IFR tools over the fictional Contoso Investments firm from the guided demo, with synthetic data only.

Can an agent change data or start calculations through MCP?+

No. Every tool is read-only. Uploads, calculations, approvals and settings stay with the people who use the portal or the Everix screens.

Which MCP clients work?+

Any client that supports remote MCP servers over streamable HTTP and can send a static Authorization header. The servers speak protocol revision 2026-07-28 and serve 2025-11-25 to older clients. Both are tested with the official MCP TypeScript client.

Where does the data go?+

Tool results go to the agent that called them, and from there to whichever model provider the agent uses. That is why keys are scoped, results are capped and every call is recorded.

Can an agent act as a named user?+

No. The credential is a key: an organization agent key in Everix Cloud, a service-account key on installed Everix. The audit trail shows that key.

Related on everix.io

Want to connect your agent?

Tell us which results your agents should read. We help you create the first key and connect your first client.