MCP servers | ATX AnalyticsTable Docs

MCP serversRegister BigQuery, PostgreSQL, Jira, Confluence or any other Model Context Protocol server from a template, test it, inspect and narrow its tools, and manage it over HTTP. Settings → MCP servers & API registers Model Context Protocol (MCP) servers: external tools that Ask Greg and the API can use. Templates are ready for Google BigQuery and PostgreSQL (through Google's MCP Toolbox for Databases), Jira and Confluence Cloud (through Atlassian's Rovo MCP server), and Jira and Confluence Server/Data Center (through the open-source mcp-atlassian). A custom server can be added as a local process (stdio) or a remote one (streamable HTTP).

MCP servers — BigQuery, PostgreSQL, Jira, Confluence and more

Settings → MCP servers & API registers Model Context Protocol (MCP) servers: external tools that Ask Greg and the API can use. Templates are ready for Google BigQuery and PostgreSQL (through Google's MCP Toolbox for Databases), Jira and Confluence Cloud (through Atlassian's Rovo MCP server), and Jira and Confluence Server/Data Center (through the open-source mcp-atlassian). A custom server can be added as a local process (stdio) or a remote one (streamable HTTP).
Pick a template, fill in the blanks (a project id, a host, a URL) and the secret values, press Test to connect and list the tools, then Save. Secrets are typed once into password fields and kept in the encrypted vault; they are never shown again or written to any log. Only an administrator can add, change, test, reset or delete a server. Everyone can see the servers and their tools.
This page is about ATX AnalyticsTable calling other MCP servers. To let an outside agent (Claude, Copilot, Cursor) call ATX AnalyticsTable for governed metrics, see Governed metrics, watches and embedding → ATX AnalyticsTable as an MCP server (signed-in documentation).
Open the MCP servers screen: Settings → Settings → Set / Restore → MCP servers & API
The screen lists every registered server with its transport, its health and whether Ask Greg may use it, and offers Tools, Edit, Reset and Delete for each. Everyone can see the list; only an administrator can add or change a server. Servers delivered by machine policy are marked "managed by policy" and cannot be changed here.

Add a server

1. Open Settings → Set / Restore → MCP servers & API.
2. Under Add from template, choose a template — Google BigQuery, PostgreSQL, Jira Cloud, Confluence Cloud, Jira or Confluence Server / Data Center, or Custom.
3. Read "Before you start": the template's prerequisites and a link to the server's own documentation.
4. Fill in what the template asks for — a project id, a host, a URL. Put passwords and Authorization headers only in the secret fields: they go to the encrypted vault and are never shown again.
5. Press Test. The result names the server, its protocol version, how long it took and how many tools it offers; a failure shows the last lines the server printed, with secrets removed.
6. Save. Then press Tools to see every tool with its class (read, SQL checked read-only, write) and whether it is allowed.
7. To let Ask Greg use the server, turn on Available to Ask Greg — and see "Ask Greg with MCP servers" for the second switch.

Templates

Google BigQuery (MCP Toolbox) — Local process: npx -y @toolbox-sdk/server --prebuilt bigquery --stdio — BIGQUERY_PROJECT (required); GOOGLE_APPLICATION_CREDENTIALS = the PATH of a service-account key file, or empty for Application Default Credentials; BIGQUERY_LOCATION; optional row limit, cost cap and write mode — list_dataset_ids, list_table_ids, get_dataset_info, get_table_info, search_catalog, execute_sql (checked read-only)
PostgreSQL (MCP Toolbox) — Local process: npx -y @toolbox-sdk/server --prebuilt postgres --stdio — POSTGRES_HOST, POSTGRES_PORT, POSTGRES_DATABASE, POSTGRES_USER, optional POSTGRES_QUERY_PARAMS; secret POSTGRES_PASSWORD — list_tables, execute_sql (checked read-only)
Jira Cloud (Atlassian Rovo MCP) — Remote: https://mcp.atlassian.com/v2/mcp — Secret Authorization: Basic <base64 of email:api_token> for a person, or Bearer <api key> for a service account — Jira search and fetch tools; write tools refused
Confluence Cloud (Atlassian Rovo MCP) — Remote: the same server — The same credential — Confluence search and fetch tools; write tools refused
Jira Server / Data Center — Local process: uvx mcp-atlassian with READ_ONLY_MODE=true — JIRA_URL; secret JIRA_PERSONAL_TOKEN — jira_* read tools
Confluence Server / Data Center — Local process: uvx mcp-atlassian with READ_ONLY_MODE=true — CONFLUENCE_URL; secret CONFLUENCE_PERSONAL_TOKEN — confluence_* read tools
Custom (local process or remote) — Either — Everything — Every tool that passes the read-only rule. Ask Greg access starts off until you have reviewed the tools.
Every parameter of every template — required or optional, default, example and meaning — is on the page MCP parameters and settings, together with a step-by-step BigQuery service-account setup.
Atlassian Cloud: an organisation administrator must enable API-token authentication for the Rovo MCP server. Some tools need OAuth scopes an API token cannot have, so a token may list fewer tools.
The MCP Toolbox templates need Node.js on the PATH of the account the backend runs as — or download the toolbox binary and point Command at it.
The Data Center templates need uv installed.
Give a database server a login that can only SELECT. The read-only check here is a second line of defence, not the first.
Any MCP server can be added today from a Custom template. A new built-in template is one entry in the product's template list, read by the screen, the API and the template test alike.

Server fields

Name — The name shown in the list, in the audit trail and to Ask Greg.
Id — Lower-case letters, digits and hyphens. Used in the API and cannot be changed later.
Description — Shown in the list, and to Ask Greg when it chooses a server.
Transport — A local process is started on this machine (npx, uvx, a binary) and spoken to over stdin and stdout. A remote server is reached over HTTPS through the installation's egress guard and proxy. — Local process (stdio); Remote (streamable HTTP)
Command — Local process only. The executable to start; an administrator can limit which ones are allowed.
Arguments — One argument per line.
Environment — KEY=value, one per line. Not for secrets.
URL — Remote only. The server's endpoint. https is required except for loopback or a private network you allow.
Headers — Name=value, one per line. Credentials go under secret names.
Allow private network — For a server inside the company network. Otherwise only public hosts are reached.
Secret names — Comma-separated variables or headers whose values are secret (stored in the vault).
Allowed tools — Comma-separated. Empty means every tool that passes the rules below.
Tool name pattern — Regular expression, e.g. ^jira_ to offer only Jira tools.
SQL arguments — Arguments that carry SQL (e.g. sql). Their value must be one read-only statement.
Read-only — Refuse tools that create, change or delete anything.
Available to Ask Greg — Also needs "Let the agent use MCP servers" in Ask Greg → Data policy.
Enabled — A disabled server is kept but never started or called.
Time limit per call (seconds) — How long one tool call may take before it is cancelled. Default 30 seconds, at most 300.
Calls at the same time — Calls this server may run at once (default 4). Further calls wait in a short queue, then are refused, so one slow server cannot hold every request.

Status

Not started — Registered but not called yet, or stopped after 10 idle minutes. The next call starts it.
Healthy — Connected; calls go through.
Paused after errors — Five calls in a row failed, so calls are refused at once for 30 seconds instead of each waiting for a timeout. Fix the cause and press Reset.
Recovering — The pause is over; one trial call decides whether the server is healthy again.
Disabled — Turned off in its settings.

Managing servers over HTTP

Everything the screen does is also an HTTP call, for configuration as code. Changing a server needs the admin role; calling a tool needs analyst.
GET — /api/mcp/templates · /api/v1/mcp/templates — viewer
GET — /api/mcp/status — viewer
GET — /api/mcp/servers · /api/v1/mcp/servers — viewer
PUT — /api/mcp/servers/{id} — body { server, secrets?, expectedUpdatedAt? } — admin
DELETE — /api/mcp/servers/{id} — admin
POST — /api/mcp/servers/{id}/test · /api/mcp/test (an unsaved form) — admin
POST — /api/mcp/servers/{id}/reset — admin
GET — /api/mcp/servers/{id}/tools · /api/v1/mcp/servers/{id}/tools — viewer
POST — /api/mcp/servers/{id}/tools/{tool}/call · /api/v1/… — body { arguments } — analyst
secrets on PUT: a key with a value replaces the stored one, a key with "" removes it, and a key left out keeps what is stored — so an edit can be saved without retyping a password nobody is shown. expectedUpdatedAt stops two administrators overwriting each other.
curl -sS -X PUT "$ATX_URL/api/mcp/servers/finance-pg" \ -H "Authorization: Bearer $ADMIN_TOKEN" -H "Content-Type: application/json" -d '{ "server": { "name": "Finance PostgreSQL", "templateId": "postgres", "transport": "stdio", "command": "npx", "args": ["-y","@toolbox-sdk/server","--prebuilt","postgres","--stdio"], "env": { "POSTGRES_HOST": "pg.corp.local", "POSTGRES_PORT": "5432", "POSTGRES_DATABASE": "finance", "POSTGRES_USER": "atx_reader" }, "secretKeys": ["POSTGRES_PASSWORD"], "allowedTools": ["list_tables","execute_sql"], "sqlArgs": ["sql"], "readOnly": true, "allowAgent": true, "timeoutMs": 60000 }, "secrets": { "POSTGRES_PASSWORD": "..." } }' # Call a tool as an integration would curl -sS -X POST "$ATX_URL/api/v1/mcp/servers/finance-pg/tools/execute_sql/call" \ -H "Authorization: Bearer $ATX_TOKEN" -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{"arguments":{"sql":"SELECT status, COUNT(*) FROM orders GROUP BY status"}}'

Protocol

Two transports are supported: a local process (newline-delimited JSON-RPC over stdin and stdout) and streamable HTTP (a POST per message; the answer may be JSON or an event stream; session and protocol-version headers are handled). Protocol revisions 2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05 are spoken. A server that only offers the older HTTP+SSE transport can be reached through the mcp-remote bridge as a local process. Requests a server makes of the client are answered — ping is answered, sampling, roots and elicitation are declined — so a server is never handed control.

ATX Institute on LinkedIn

This site needs JavaScript for the full experience. Every page above is reachable without it.