MCP parameters and settings | ATX AnalyticsTable Docs

MCP parameters and settingsEvery value you fill in for an MCP server — per template, with what is required, defaults, examples and meaning — how the Environment, Secret and Header boxes work, connecting BigQuery with a service account, and every installation setting for the API and MCP. When you choose a template, the form is pre-filled and a Parameters table explains every value: its name, where it goes (Environment, Secret or Header), whether it is required, an example and what it means. The Environment box holds one KEY=value per line. Each line becomes an environment variable of the server process, and an empty value (KEY=) means NOT SET, so the server uses its own default. The Environment box is stored as plain text, so never put a password or key in it.

Setting up an MCP server — the parameters

When you choose a template, the form is pre-filled and a Parameters table explains every value: its name, where it goes (Environment, Secret or Header), whether it is required, an example and what it means. The Environment box holds one KEY=value per line. Each line becomes an environment variable of the server process, and an empty value (KEY=) means NOT SET, so the server uses its own default. The Environment box is stored as plain text, so never put a password or key in it.
A password goes under Secret names: type its name there (for example POSTGRES_PASSWORD), and a password field appears; the value is kept in the encrypted vault and never shown again. A key FILE, such as a Google service-account key, is never pasted: save it as a file on the machine that runs ATX AnalyticsTable and enter its path. Saving is refused if a private key is found in the Environment or Headers box. Test checks the parameters BEFORE it starts anything.
A required value that is empty, a path that does not exist, a key pasted instead of a path, a number that is not a number or a value outside the allowed choices is reported in plain words. When you edit a server saved before a template gained new parameters, Add the missing lines adds them with their default values.
Open the MCP servers screen: Settings → Settings → Set / Restore → MCP servers & API

How the boxes work

Environment (local servers) — One KEY=value per line. Each line becomes an environment variable of the server process. KEY= with nothing after it means not set: the server uses its own default. — Plain text in the server list. A private key found here is refused on save.
Headers (remote servers) — One Name=value per line, sent with every request. — Plain text. A credential header (Authorization, X-Api-Key, Cookie…) must be listed under Secret names instead.
Secret names — Comma-separated names of variables or headers whose values are secret. Each name gets a password field. — The encrypted credential vault. Never shown again, never logged, never in the audit trail.
A key FILE (e.g. a Google service-account key) — Never pasted. Save it as a file on the machine that runs ATX AnalyticsTable and put its PATH in the Environment box. — The file stays where you put it; protect it with file permissions.
Test checks the parameters before it starts anything, and says what is wrong in plain words: a required value that is empty, a path that does not exist on this machine (or that the account running ATX AnalyticsTable cannot see), a key pasted instead of a path, a number that is not a number, or a value outside the allowed choices. The same check runs before a saved server is started for Tools, the API or Ask Greg.
Editing a server saved before its template gained new parameters shows "Add the missing lines": it appends the missing Environment lines with the template's values, for example BIGQUERY_LOCATION= and BIGQUERY_WRITE_MODE=blocked on an older BigQuery server.

Connecting BigQuery with a service account

1. In Google Cloud Console open IAM → Service accounts and create an account, for example atx-reader.
2. Grant it BigQuery Data Viewer on the datasets it may read and BigQuery Job User on the project. Nothing that can write.
3. Open the account → Keys → Add key → JSON. A key file is downloaded.
4. Move the file to a folder only the account that runs ATX AnalyticsTable can read, for example C:\Keys\atx-bigquery-sa.json, and restrict it (command below).
5. In ATX AnalyticsTable open Settings → Set / Restore → MCP servers & API → Add from template → Google BigQuery (MCP Toolbox).
6. Fill in the Environment box as in the example below: BIGQUERY_PROJECT is the project_id from the key file, GOOGLE_APPLICATION_CREDENTIALS is the PATH of the key file, BIGQUERY_LOCATION is where your datasets live.
7. Press Test. The first test downloads MCP Toolbox through npx and can take a minute; it should report about nine tools. Then Save and press Tools: six tools are allowed and execute_sql is SQL checked read-only.
8. To let Ask Greg use it, keep Available to Ask Greg on and turn on Let the agent use MCP servers in Ask Greg → Governance → Data policy (see Ask Greg with MCP servers).
icacls C:\Keys\atx-bigquery-sa.json /inheritance:r /grant:r "%USERNAME%:(R)"
BIGQUERY_PROJECT=my-analytics-project GOOGLE_APPLICATION_CREDENTIALS=C:\Keys\atx-bigquery-sa.json BIGQUERY_LOCATION=EU BIGQUERY_MAX_QUERY_RESULT_ROWS=50 BIGQUERY_MAXIMUM_BYTES_BILLED=10737418240 BIGQUERY_WRITE_MODE=blocked
The two mistakes that stop BigQuery from starting: BIGQUERY_PROJECT left empty, and the service-account JSON pasted into GOOGLE_APPLICATION_CREDENTIALS instead of the file's path. Both are now reported by Test before anything runs; saving a pasted private key is refused.
The other lines are optional. BIGQUERY_MAX_QUERY_RESULT_ROWS is the most rows one query returns (default 50); BIGQUERY_MAXIMUM_BYTES_BILLED is a cost cap per query in bytes (10737418240 = 10 GiB; empty = no cap); BIGQUERY_WRITE_MODE=blocked makes MCP Toolbox itself refuse anything but SELECT; BIGQUERY_IMPERSONATE_SERVICE_ACCOUNT runs the queries as another service account.
Without a key file: run gcloud auth application-default login once on the machine that runs ATX AnalyticsTable and leave GOOGLE_APPLICATION_CREDENTIALS empty. The Toolbox then uses that sign-in. On a Google Cloud VM, the VM's attached service account is used the same way.
Watch it done: the training film Ask Greg about BigQuery (1:19, 23 steps, recorded on the real application with a real BigQuery server) — AnalyticsTable → Training, /products/analyticstable/trainings.

PostgreSQL, Jira and Confluence — what to fill in

PostgreSQL (MCP Toolbox): POSTGRES_HOST, POSTGRES_DATABASE and POSTGRES_USER are required. POSTGRES_PORT defaults to 5432. POSTGRES_PASSWORD is a secret. POSTGRES_QUERY_PARAMS adds connection settings such as sslmode=require. Use a login with SELECT privileges only. Jira Cloud and Confluence Cloud (Atlassian Rovo MCP) need one secret header, Authorization. For a person use Basic followed by the base64 of e-mail:API-token. For an Atlassian service account use Bearer followed by the API key.
An organisation admin must first enable API-token authentication for the Rovo MCP server. Jira and Confluence Server / Data Center (mcp-atlassian) need JIRA_URL or CONFLUENCE_URL and the secret JIRA_PERSONAL_TOKEN or CONFLUENCE_PERSONAL_TOKEN. READ_ONLY_MODE=true makes the server refuse writes itself. JIRA_SSL_VERIFY and CONFLUENCE_SSL_VERIFY check the TLS certificate. JIRA_PROJECTS_FILTER and CONFLUENCE_SPACES_FILTER limit searches to the listed keys.
These templates need uv installed. A custom server has no parameter table: its own documentation says which variables it reads.

Template reference

Generated from the templates the application ships, so it always matches the Parameters table under the form in Settings. "Template" in the Pre-filled column is the value the form starts with; "default" is what the server uses when the line is empty.
Google BigQuery (MCP Toolbox)
Datasets, tables, schema and read-only SQL on BigQuery through Google's MCP Toolbox for Databases. Runs as a local process: npx -y @toolbox-sdk/server --prebuilt bigquery --stdio. Template id: bigquery. Ask Greg access starts on; read-only on; time limit 60 s per call.
BIGQUERY_PROJECT — Environment — required — — — my-analytics-project — The Google Cloud project id the tools run in: queries are billed to it and a table name without a project is looked up in it. It is the project_id field of a service-account key file.
GOOGLE_APPLICATION_CREDENTIALS — Environment — optional — — — C:\Keys\atx-bigquery-sa.json — The PATH of a service-account key file (JSON) on the machine that runs ATX AnalyticsTable — never the JSON itself. Leave it empty to use Application Default Credentials: gcloud auth application-default login on this machine, or the service account attached to a Google Cloud VM.
BIGQUERY_LOCATION — Environment — optional — — — EU — Where query jobs run: EU, US or a region such as europe-west3. Set it to the location of your datasets; left empty, BigQuery works it out.
BIGQUERY_MAX_QUERY_RESULT_ROWS — Environment — optional — template: 50; default: 50 — 50 — The most rows one execute_sql call returns. Aggregated questions fit well within 50; the data policy still decides how much of a result Ask Greg is shown.
BIGQUERY_MAXIMUM_BYTES_BILLED — Environment — optional — default: 0 — 10737418240 — A cost cap per query, in bytes: BigQuery refuses, without running it, a query that would scan more. 10737418240 is 10 GiB. 0 or empty means no cap.
BIGQUERY_WRITE_MODE — Environment — optional — template: blocked — blocked — MCP Toolbox's own write protection. blocked lets only SELECT statements through; the template sets it as a second wall behind ATX AnalyticsTable's read-only check. Allowed: blocked, protected, allowed.
BIGQUERY_IMPERSONATE_SERVICE_ACCOUNT — Environment — optional — — — atx-reader@my-analytics-project.iam.gserviceaccount.com — Run as this service account by impersonation instead of as the credentials' own identity. The identity behind the credentials needs the Service Account Token Creator role on it.
Tools offered by default: list_dataset_ids, list_table_ids, get_dataset_info, get_table_info, execute_sql, search_catalog. SQL arguments checked read-only: sql.
Needs Node.js (npx) on this machine, or replace the command with the path of a downloaded toolbox binary and keep the arguments after the package name.
BIGQUERY_PROJECT is required: the project_id in your service-account key file.
GOOGLE_APPLICATION_CREDENTIALS is the PATH of the key file on this machine (e.g. C:\Keys\atx-bigquery-sa.json), not the JSON itself. Leave it empty to use gcloud auth application-default login.
Give the service account BigQuery Data Viewer on the datasets and BigQuery Job User on the project — nothing that can write.
An empty line means "not set": the Toolbox then uses its default.
execute_sql is offered only because its SQL is checked to be one read-only statement before it is sent.
Vendor documentation: https://docs.cloud.google.com/bigquery/docs/pre-built-tools-with-mcp-toolbox
PostgreSQL (MCP Toolbox)
Tables and read-only SQL on a PostgreSQL database (on-premises, Cloud SQL, AlloyDB, RDS, Azure) through MCP Toolbox. Runs as a local process: npx -y @toolbox-sdk/server --prebuilt postgres --stdio. Template id: postgres. Ask Greg access starts on; read-only on; time limit 60 s per call.
POSTGRES_HOST — Environment — required — — — pg.corp.local — Host name or IP address of the PostgreSQL server (Cloud SQL, AlloyDB, RDS and Azure included).
POSTGRES_PORT — Environment — optional — template: 5432; default: 5432 — 5432 — TCP port of the server.
POSTGRES_DATABASE — Environment — required — — — finance — The database to connect to.
POSTGRES_USER — Environment — required — — — atx_reader — The database login. Give it SELECT privileges only.
POSTGRES_PASSWORD — Secret (vault) — required — — — — — The password of that login. Kept in the encrypted vault and handed to the process at start; never shown again.
POSTGRES_QUERY_PARAMS — Environment — optional — — — sslmode=require — Extra connection parameters added to the connection string, for example sslmode=require for TLS.
Tools offered by default: list_tables, execute_sql. SQL arguments checked read-only: sql.
Use a database role with SELECT privileges only — the read-only check here is a second line of defence, not a replacement for database permissions.
The password is stored in the encrypted vault and given to the process as POSTGRES_PASSWORD; it is never shown again.
Vendor documentation: https://github.com/googleapis/mcp-toolbox
Jira Cloud (Atlassian Rovo MCP)
Search Jira issues with JQL and read issues, projects and comments through Atlassian's hosted MCP server. Reached over streamable HTTP at https://mcp.atlassian.com/v2/mcp. Template id: jira-cloud. Ask Greg access starts on; read-only on; time limit 45 s per call.
Authorization — Header (secret) — required — — — Basic <base64 of you@example.com:api_token> — Listed under Secret names, so it is kept in the vault. Basic with the base64 of e-mail:API-token for a person, or Bearer <API key> for an Atlassian service account. An organisation admin must first enable API-token authentication for the Rovo MCP server.
Tools offered: those whose name matches the pattern jira|search|fetch|getAccessibleAtlassianResources|atlassianUserInfo|lookupJiraAccountId.
An Atlassian organisation admin must enable API-token authentication for the Rovo MCP server first.
Authorization is either Basic <base64 of email:api_token> for a personal token or Bearer <api key> for a service account.
Write tools (create/edit/transition issue, add comment) are refused while Read-only is on.
Vendor documentation: https://support.atlassian.com/atlassian-rovo-mcp-server/docs/configuring-authentication-via-api-token
Confluence Cloud (Atlassian Rovo MCP)
Search Confluence with CQL and read spaces and pages through Atlassian's hosted MCP server. Reached over streamable HTTP at https://mcp.atlassian.com/v2/mcp. Template id: confluence-cloud. Ask Greg access starts on; read-only on; time limit 45 s per call.
Authorization — Header (secret) — required — — — Basic <base64 of you@example.com:api_token> — Listed under Secret names, so it is kept in the vault. Basic with the base64 of e-mail:API-token for a person, or Bearer <API key> for an Atlassian service account. An organisation admin must first enable API-token authentication for the Rovo MCP server.
Tools offered: those whose name matches the pattern confluence|search|fetch|getAccessibleAtlassianResources|atlassianUserInfo.
Same server and credential as Jira Cloud; the tool pattern picks the Confluence tools.
Page bodies can be long — results are cut to the per-call size limit before they reach Ask Greg.
Vendor documentation: https://support.atlassian.com/atlassian-rovo-mcp-server/docs/configuring-authentication-via-api-token
Jira Server / Data Center (mcp-atlassian)
Self-hosted Jira through the open-source mcp-atlassian server, started locally with uvx, in read-only mode. Runs as a local process: uvx mcp-atlassian. Template id: jira-datacenter. Ask Greg access starts on; read-only on; time limit 45 s per call.
JIRA_URL — Environment — required — — — https://jira.corp.local — Base address of your Jira Server or Data Center.
JIRA_PERSONAL_TOKEN — Secret (vault) — required — — — — — A personal access token created in your profile there. Kept in the vault.
READ_ONLY_MODE — Environment — optional — template: true; default: true — true — Makes mcp-atlassian itself refuse every write. Keep it true. Allowed: true, false.
JIRA_SSL_VERIFY — Environment — optional — template: true; default: true — true — Check the server's TLS certificate. Set false only for a test server with a self-signed certificate. Allowed: true, false.
JIRA_PROJECTS_FILTER — Environment — optional — — — FIN,OPS — Comma-separated project keys to limit searches to.
Tools offered: those whose name matches the pattern ^jira_.
Needs uv (uvx) installed. READ_ONLY_MODE=true makes the server itself refuse writes as well.
Vendor documentation: https://github.com/sooperset/mcp-atlassian
Confluence Server / Data Center (mcp-atlassian)
Self-hosted Confluence through mcp-atlassian, started locally with uvx, in read-only mode. Runs as a local process: uvx mcp-atlassian. Template id: confluence-datacenter. Ask Greg access starts on; read-only on; time limit 45 s per call.
CONFLUENCE_URL — Environment — required — — — https://confluence.corp.local — Base address of your Confluence Server or Data Center.
CONFLUENCE_PERSONAL_TOKEN — Secret (vault) — required — — — — — A personal access token created in your profile there. Kept in the vault.
READ_ONLY_MODE — Environment — optional — template: true; default: true — true — Makes mcp-atlassian itself refuse every write. Keep it true. Allowed: true, false.
CONFLUENCE_SSL_VERIFY — Environment — optional — template: true; default: true — true — Check the server's TLS certificate. Set false only for a test server with a self-signed certificate. Allowed: true, false.
CONFLUENCE_SPACES_FILTER — Environment — optional — — — FIN,POLICY — Comma-separated space keys to limit searches to.
Tools offered: those whose name matches the pattern ^confluence_.
Needs uv (uvx) installed. READ_ONLY_MODE=true makes the server itself refuse writes as well.
Vendor documentation: https://github.com/sooperset/mcp-atlassian
Custom MCP server (local process)
Any MCP server started as a local process speaking JSON-RPC over stdin/stdout. Runs as a local process: (your command). Template id: custom-stdio. Ask Greg access starts off; read-only on; time limit 30 s per call.
No parameter table: a custom server reads whatever variables its own documentation lists. Put ordinary values in the Environment box and passwords under Secret names.
Starts with Ask Greg access OFF: review the tool list first, then turn it on.
Vendor documentation: https://modelcontextprotocol.io/specification
Custom MCP server (streamable HTTP)
Any remote MCP server that speaks the streamable HTTP transport. Reached over streamable HTTP. Template id: custom-http. Ask Greg access starts off; read-only on; time limit 30 s per call.
No parameter table: a custom server reads whatever variables its own documentation lists. Put ordinary values in the Environment box and passwords under Secret names.
Servers that only speak the legacy HTTP+SSE transport can be reached through the mcp-remote bridge as a local process.
Vendor documentation: https://modelcontextprotocol.io/specification

Installation settings for the API and MCP

These are set on the machine or server that runs ATX AnalyticsTable — as environment variables of the service, or, where a config.ini key is given, in config.ini. Like every switch in the product, they can only narrow what the Settings screen allows. A change takes effect when the backend restarts, except the tokens file, which is re-read when it changes.
ANALYTICS_PUBLIC_API — — — on — 0 turns /api/v1 off; every call answers 503 disabled.
ANALYTICS_API_TOKENS_FILE — [rbac] tokens_file — — — The scoped service tokens file: a JSON list of { name, sha256, role }. Re-read when it changes.
ANALYTICS_API_RATE_PER_MIN — — — 600 — Requests a minute per credential.
ANALYTICS_API_BURST — — — 100 — Bucket size per credential.
ANALYTICS_MCP — — — on — 0 turns every MCP server off; the Settings screen says so and MCP routes answer 503 disabled.
ANALYTICS_MCP_STDIO — — — on — 0 allows remote (HTTP) MCP servers only — no local processes. The usual choice for a shared server.
ANALYTICS_MCP_ALLOWED_COMMANDS — — — any — Comma-separated executables a local server may be started with, e.g. npx,uvx,toolbox.
ANALYTICS_MCP_MAX_RESULT_KB — — — 256 — Largest tool result kept; longer text is cut and marked truncated.
ANALYTICS_MCP_SERVERS_FILE — — — — — MCP servers delivered by machine policy, as { "servers": [ … ] }. Shown as managed and read-only in Settings; their secrets are still entered once.
AI_ALLOW_MCP — — — — — 0: Ask Greg may never use MCP, whatever the data policy says. No value turns it on.
ANALYTICS_MCP_SERVER — — — off — 1 lets outside agents use ATX AnalyticsTable itself as an MCP server for governed metrics at POST /api/v1/mcp (the other direction from the servers on this page). Needs the mcp-server licence component where licensing is enforced.
ANALYTICS_MCP_SERVER_MAX_ROWS — — — 500 — Rows one query_metrics call of that MCP server returns at most.
Where the server list itself lives: %APPDATA%\.analyticstable\mcp\servers.json on Windows (~/.analyticstable/mcp/servers.json elsewhere, or under ANALYTICS_USER_DATA). It never contains a secret: those are in the vault under mcp.<server id>.
{ "servers": [ { "id": "finance-bq", "name": "Finance BigQuery", "templateId": "bigquery", "transport": "stdio", "command": "npx", "args": ["-y", "@toolbox-sdk/server", "--prebuilt", "bigquery", "--stdio"], "env": { "BIGQUERY_PROJECT": "finance-prod", "GOOGLE_APPLICATION_CREDENTIALS": "C:\\ProgramData\\AnalyticsTable\\keys\\bq.json", "BIGQUERY_LOCATION": "EU", "BIGQUERY_WRITE_MODE": "blocked" }, "allowedTools": ["list_dataset_ids", "list_table_ids", "get_table_info", "execute_sql"], "sqlArgs": ["sql"], "readOnly": true, "allowAgent": true } ] }
The example above is a machine-policy file for ANALYTICS_MCP_SERVERS_FILE: the same fields as the Settings form. In JSON, every backslash of a Windows path is written twice.

ATX Institute on LinkedIn

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