Vendor MCP servers

Some vendors run an official MCP server. Connect one as a Custom server under Connections → Add connection. SecondShift has no built-in support for these vendors. Any MCP server that offers sign-in or a bearer token connects the same way.

Connect any MCP server

1. Find the vendor’s MCP server URL. It usually ends in /mcp. Use a read-only URL if the vendor offers one.
2. Choose Sign in if the server offers it. Tick only the scopes the job needs, and prefer read scopes. SecondShift refuses the sign-in if the server grants more than you ticked. Without sign-in, use the narrowest bearer token the vendor offers.
3. Sign in as an account that can see only what jobs should see. The credential sets the reach. Unselected tools do not limit it.
4. Allow only the tools jobs need. Tools the server marks read-only start selected. May change data means the server did not mark the tool read-only. Read the tool’s description before allowing it.
5. Add a requires block with the binding and the tools. Run the job once by hand and read the run’s timeline before scheduling it.

Example: Linear

Checked on 2026-11-01. The tools and requires block below are a sample for a job that reads issues and projects. Pick the tools your own job needs.

Setting	Value
Binding	linear
Server URL	https://mcp.linear.app/mcp/readonly
Sign-in	Sign in, with the read scope ticked

Linear’s /mcp/readonly endpoint offers only read tools, and the read scope keeps the token away from Linear’s write APIs. Use the plain /mcp endpoint only for a job that must create or change issues.

Linear marks none of its tools read-only. No tool starts selected, and every tool shows May change data. On this endpoint with the read scope, none of them can change data.

This example allows these tools on the connection:

Tool	Returns
list_teams	Teams in the workspace
list_issues	Issues, filtered by team, state, assignee, label, project or date
get_issue	One issue with its attachments and branch name
list_comments	Comments on an issue, project or document
list_projects	Projects in the workspace
get_project	One project, with its milestones and members on request

requires:
  mcps:
    - binding: linear
      tools:
        - list_teams
        - list_issues
        - get_issue
        - list_comments
        - list_projects
        - get_project

list_issues skips archived issues unless the job passes includeArchived.

What the credential reaches. The token acts as the person who signed in, in the Linear workspace they signed in to. It can read everything that person can see there, including private teams they belong to. Other tools on this endpoint read that person’s notifications (get_notifications), list members with their email addresses (list_users, get_user), and fetch images from issue text (extract_images). Allow them only for a job that needs them.

Example: Grafana

Checked on 2026-11-01 with mcp-grafana 2.1.1, Grafana 12.2, Prometheus and Loki. The tools and requires block below are a sample for a job that triages alerts. Pick the tools your own job needs.

Grafana’s MCP server, mcp-grafana, runs next to your Grafana. It holds the Grafana credential. SecondShift holds a separate token that only lets it call the server.

Setting	Value
Binding	grafana
Server URL	Where you run mcp-grafana, ending in /mcp
Sign-in	Bearer token, set to the server’s --server-auth-token

Run the server

1. In Grafana, add a service account with the Viewer role, and create a token for it. Viewer can read dashboards, alert rules and data sources, and cannot change them.

2. Run mcp-grafana with that token and a second, random token for callers:

   docker run -d -p 8000:8000 \
     -e GRAFANA_URL=https://grafana.example.com \
     -e GRAFANA_SERVICE_ACCOUNT_TOKEN=<service account token> \
     grafana/mcp-grafana:2.1.1 \
     -t streamable-http --address 0.0.0.0:8000 \
     --disable-write \
     --server-auth-token <random token> \
     --allowed-hosts mcp-grafana.example.com \
     --loki-guardrail-mode enforce --loki-guardrail-max-range 24h

   • --disable-write removes the tools that change Grafana.
   • --server-auth-token refuses callers without it. Without it, anyone who can reach the server uses the service account.
   • --allowed-hosts lists the host names callers use. The server refuses others, so include the one in the URL SecondShift calls.
   • The Loki flags refuse a log query over 24 hours. The run sees a failed tool call, and the server logs why. A query returns at most 100 lines, set with --max-loki-log-limit. Version 2.1.1 has no matching limit for Prometheus queries. Prometheus’s own query limits still apply.

3. Serve it over HTTPS. Outside local mode, SecondShift calls only HTTPS URLs.

Connect it

Add a Custom server connection with the URL, such as https://mcp-grafana.example.com/mcp, and the random token as the bearer token. With --disable-write, the server marks every tool it offers read-only, so every tool starts selected. Clear the ones jobs do not need.

This example allows these tools. A run read a Prometheus range and Loki error lines for a 30 minute window with them.

Tool	Returns
list_datasources	Data sources and their IDs, which the query tools take
alerting_rules_read	Alert rules, with their queries and state
query_prometheus	A PromQL query, instant or over a range
list_prometheus_metric_names	Metric names, to find the right one
query_loki_logs	Log lines from a LogQL query
list_loki_label_names	Log labels, to build a query
search_dashboards	Dashboards by title

requires:
  mcps:
    - binding: grafana
      required: false
      tools:
        - list_datasources
        - alerting_rules_read
        - query_prometheus
        - list_prometheus_metric_names
        - query_loki_logs
        - list_loki_label_names
        - search_dashboards

required: false lets the job run, and say Grafana was not checked, when the connection is missing.

What the credential reaches. The service account reads every dashboard, alert rule and data source a Viewer can see in its organization. Through the data sources, that is all their metrics and logs. Other tools reach more:

• grafana_api_request calls any Grafana API path. The server marks it read-only, but it is not limited to reads.
• search_docs and get_doc fetch pages from the internet.
• The OnCall, Incident and Sift tools read people’s schedules and incidents.

Allow them only for a job that needs them.

Start a job from a Grafana alert

A job with triggers: [webhook] can start when a Grafana alert fires. See webhook triggers for how triggers work.

1. Add the trigger and these settings to the job, and sync the repository:

   triggers: [webhook]
   webhook:
     skip_when:
       status: resolved
     repeat_key: groupKey
     repeat_window: 1h

   Grafana sends status: resolved when an alert stops firing, and the same groupKey each time it repeats a notification for the same group.

2. On the job page, turn on the webhook trigger. Copy the URL and the secret.

3. In Grafana, add a contact point with the Webhook integration and that URL. Set Authorization Header - Scheme to Bearer and Authorization Header - Credentials to the secret.

4. Add a notification policy that routes the alerts to it. Keep an existing Slack contact point on the same policy, so people still get the alert.

5. Use Test on the contact point. The job page lists the request.

To sign requests instead of sending the secret, leave the Authorization header empty. Under HMAC Signature, set Secret to the secret, Header to X-SecondShift-Signature and Timestamp header to X-SecondShift-Timestamp.

Grafana must reach the trigger URL. A Grafana in Docker reaches a SecondShift on the same machine at host.docker.internal. Add that name to SECONDSHIFT_ALLOWED_HOSTS, or SecondShift answers 400.

Other vendors

On 2026-11-01, SecondShift found sign-in settings on these servers. Nobody completed a sign-in or ran a job against them. Follow the steps above and check the tool list after connecting.

Vendor	Server URL	Sign-in	Limiting access
Atlassian	https://mcp.atlassian.com/v2/mcp?tools=all	Offered	Tick read scopes only, such as read:jira:agent-interface and read:confluence:agent-interface. Keep ?tools=all. Without it, the server offers generic discover and execute tools instead of named ones.
Datadog	https://mcp.datadoghq.com/v1/mcp. Other sites use their own host, such as mcp.datadoghq.eu.	Offered, with one scope, mcp_all. Datadog’s docs also allow a personal or service access token as a bearer token.	Pick tools with ?toolsets=. Only core is on by default. Users without the mcp_write permission get no write tools.
Notion	https://mcp.notion.com/mcp	Offered, with one scope, default	No read-only option is documented. The token reaches everything the person who signed in can see.

SecondShift cannot sign in to PagerDuty’s server (https://mcp.pagerduty.com/mcp), because it does not let new apps register. PagerDuty’s docs send API tokens as Token token=..., and SecondShift sends tokens as Bearer, so the token option was not tried.