Skip to main content
This article explains how to use the Cortex MCP. For information on setting up the Cortex MCP, refer to Configuring the Cortex MCP.
We’d love your feedback!
If you’re using the Cortex MCP, we’d love to hear your direct feedback on the configuration process, the tools available, and how you’ve been using it. Your feedback helps us improve and expand Cortex’s capabilities. If you’re interested in helping guide Cortex’s MCP direction, take the survey.

Starting a new chat

In your MCP client, ask a question about your Cortex workspace. The client uses the Cortex API to provide detailed answers based on your data. Examples
  • You might ask what’s going on with a specific Scorecard. The MCP client responds with an overview, Scorecard structure information, a progress summary, and suggestions for next steps.
    Screenshot 1 of MCP example conversation
  • You could ask about an entity’s custom data. The MCP client responds with that entity’s custom data and information about when it was created.
    Screenshot 2 of MCP example conversation
  • You could ask about an entity’s dependencies. The MCP client responds with a list of incoming and outgoing dependencies.
    Screenshot 3 of MCP example conversation

Available MCP tools

The Cortex MCP can use the following Cortex REST API endpoints. Some tools read data, and others create or update data in your workspace. The Cortex MCP can’t delete data. It runs bulk operations only on ownership recommendations. See Making changes with the Cortex MCP.
Additional tools
  • getMyWorkspace - Enabled by default, this tool supports the “Mine” capabilities that are available in the Cortex UI. Ask questions like, “show me my Cortex entities” or “tell me how my entities are performing against the Production Readiness Scorecard.”
Additional tools
  • listMetricDefinitions - This tool returns the Eng Intelligence metrics available in your workspace, along with the metadata needed to query them. Ask questions like, “which engineering metrics can I ask about?” or “what deployment data is available in Eng Intelligence?”
  • queryPointInTimeMetrics - This tool returns current values for one or more Eng Intelligence metrics, with optional comparisons against a previous period and grouping by team, entity, or user. Ask questions like, “what’s our deployment frequency this quarter compared to last?” or “show me change failure rate by team for the last 30 days.”
  • createQuery - Run a CQL query across your entities. This doesn’t change any data.
  • getQueryResults - Results of a CQL query you ran
Upload a dependency manifest or lockfile to record an entity’s packages:
  • listIntegrationStatuses - Installed integrations and the result of each one’s most recent connection check
  • query_docs - This tool queries the Cortex help center.
  • get_more_tools - This tool discovers additional specialized tools.

Limiting tools with toolsets

Some MCP clients and AI model providers limit how many tools a single request can include, and that limit counts the client’s own built-in tools too. If the full set of Cortex MCP tools goes over your client’s limit, the client may fail to connect or return an error. Toolsets let you connect to a smaller, named set of Cortex tools instead. Toolsets are available for the remote Cortex MCP.

Available toolsets

If you don’t specify a toolset, the Cortex MCP lists every available tool. The tools in each toolset can change as Cortex adds new tools.

Selecting a toolset

Add the toolsets query parameter to the end of your MCP server URL. Toolsets work whether you sign in with a personal access token or with OAuth: If your workspace is in the EU region, replace mcp.cortex.io with mcp.eu.cortex.io. With a personal access token, keep sending the Authorization: Bearer <CORTEX_TOKEN> header as usual. To use more than one toolset, separate the names with commas. The Cortex MCP lists every tool that’s in at least one of the toolsets you choose:
For example, to add the remote Cortex MCP to Claude Code with the readonly toolset:
Put quotes around the URL when you use it in a terminal command. Some shells, such as zsh, treat ? as a special character and won’t run the command if the URL isn’t quoted.
If you enter a toolset name that doesn’t exist, the Cortex MCP returns an error that names the unknown toolset and lists the valid ones.
A toolset only controls which tools your MCP client sees. It doesn’t change what your Cortex account is allowed to do. The common toolset includes tools that create and update data, so if you don’t want your MCP client to make changes, use the readonly toolset.

Enabling and disabling provided tools

Before submitting questions or commands, you may want to limit the number of provided tools being used by the MCP provider. For example, you might want to use the MCP to ask about entities, but you don’t want to allow questions about Scorecards. If you want the Cortex MCP to stay read-only, the simplest option is the readonly toolset. You can also turn off individual tools in your MCP client: turn off the tools that create, update, archive, upload, approve, or run something. Read tools generally start with get, list, search, retrieve, or query. createQuery runs a CQL query without changing data, so you can leave it on. Example In the settings of your MCP client, it’s possible to limit which Cortex tools are being used. For example, in Claude desktop:
  1. Navigate to Settings > Connectors > Cortex > 3 dots icon > Tools and Settings.
  2. Below Provided Tools, toggle off the options you don’t need.
  3. Restart Claude.
  4. Navigate to your Claude chat and enter commands.

Eng Intelligence metrics in the Cortex MCP

Eng Intelligence metrics are also available to Cortex MCP. This brings metric data directly into the Cortex MCP, enabling richer AI-assisted insights, not only into engineering performance, but across your entire Cortex ecosystem. It allows you to be a more proactive leader and enables more data-driven planning sessions.

Querying Eng Intelligence metrics in the Cortex MCP

Ensure you have the latest Cortex MCP version configured in your environment. Once enabled, the MCP automatically includes endpoints for Eng Intelligence metrics. You can then query Eng Intelligence metrics alongside ownership, Scorecard, and Initiative information. For example:
  • Evaluate my DORA metric performance in Q3. Highlight where the team is performing well and bring areas of improvement to my attention.
  • Which services have concerning MTTR trends, and what scorecard gaps are contributing?
  • Recommend initiatives and scorecard changes I could consider to address my upward trend in incidents.
  • Review last quarter’s performance against our goals, identify the top three bottlenecks, and suggest changes to our scorecards to address them.
  • Is there a correlation between my PR size and how fast we ship?

Making changes with the Cortex MCP

In addition to answering questions, the Cortex MCP can create and update data in your workspace. For example, you can ask your client to create an Initiative, record a deploy for an entity, or request a Scorecard rule exemption. See Available MCP tools for the full list.

Prerequisites

  • The Cortex MCP acts as the user who owns your personal access token. It can only make changes that user has permission to make in Cortex.
  • Make sure you’re using the latest version of the Cortex MCP. If you host it locally, pull the latest ghcr.io/cortexapps/cortex-mcp:latest image.

Example prompts

  • Create an Initiative that asks every service in the Payments group to reach Gold on the Production Readiness Scorecard by March 31, 2027.
  • Add a deploy to the payments-api entity for version 2.4.1 in the production environment.
  • Request an exemption from the “Has a runbook” rule on the Production Readiness Scorecard for legacy-billing. It’s being retired this quarter.
  • Archive the old-search entity.
  • Add tier: 1 as custom data on the checkout entity.
  • Find entities in the Payments domain without a team owner and accept Cortex’s high-confidence ownership recommendations.

What the Cortex MCP can’t do

  • Delete data. To protect your data from accidental removal, no delete endpoints are available through the Cortex MCP. This includes deleting custom data, deploys, and custom events. Delete data in Cortex or with the Cortex REST API instead. See Deleting custom data for more information.
  • Run bulk operations. Endpoints that change many records at once aren’t available, except the endpoints that accept, reject, or restore many ownership recommendations. This includes adding custom data, dependencies, or custom metric data points in bulk, verifying entities in bulk, uploading multiple Java packages at once, and replacing all team hierarchy relationships. Make these changes in Cortex or with the REST API.
  • Create teams, edit team details, or change relationships and OpenAPI docs. The Cortex MCP can’t create a team, update a team’s details, add or update an entity’s relationship sources and destinations, or update an entity’s OpenAPI documentation. Make these changes in Cortex or with the REST API.
  • Manage workspace administration. Integration configurations, API keys, users, and workspace settings aren’t available through the Cortex MCP.

Things to know

  • Some tools create a new object or replace an existing one that has the same tag: createOrUpdateEntity, createOrUpdateScorecard, createOrUpdateScaffolder, upsertCatalogPage, and updatePlugin. If you aren’t sure whether the tag is already in use, ask your client to look it up first.
  • To change only some fields on an entity or plugin, ask your client to use createOrPatchEntity or patchPlugin, which leave everything else as is.
  • Archiving is reversible. Use unarchiveEntity or unarchiveTeam to restore an archived entity or team.
  • Many MCP clients ask you to approve each tool call. Check the details before you approve a change.

Creating and running Workflows with the Cortex MCP

You can use the Cortex MCP to create, update, and run Workflows without leaving your MCP client. For example, you can ask your client to build a Workflow from a description, start a run, and then check on the run until it finishes.

Prerequisites

  • To create or update a Workflow, the user who owns your personal access token needs the Edit Workflows permission.
  • To run a Workflow, the user needs the Execute Workflow Runs permission. The Viewer role doesn’t include this permission.
  • The Workflow must be published (not a draft) and have Runnable via API turned on. See Enabling API access.
If a Workflow doesn’t meet these requirements, the MCP client tells you what’s blocking the run, so you can fix it in Cortex and try again.

Example prompts

  • Run the “Provision staging database” Workflow.
  • Run the “Rotate service credentials” Workflow for the payments-api entity, then tell me when it finishes.
  • Create a Workflow that posts a message to the #deploys Slack channel, and make it runnable via API.
  • Show me the runs of the “Onboard new service” Workflow from the last week. Did any of them fail?

Things to know

  • createOrUpdateWorkflow replaces any existing Workflow that has the same tag. Ask your client to look up the Workflow first if you aren’t sure the tag is unique.
  • When you run a Workflow for a specific entity, include the entity in your prompt. A Workflow scoped to an entity can’t run without one.
  • By default, run results returned to the MCP leave out the run context (for example, secret names and the entity descriptor) to keep responses short. To get the context, ask your client to include it.
  • listWorkflowRuns returns runs from the last 30 days.
  • You can’t delete Workflows through the Cortex MCP. Delete them in Cortex instead.

Crafting effective MCP prompts

The Cortex MCP becomes more powerful when using it becomes second nature to your teams. That shift happens when you have developed prompts that fit your specific role, answer your recurring questions, and surface information in the exact moments you need it. Refer to the Cortex MCP Prompt Library for more information.

Troubleshooting and FAQ

See frequently asked questions below.
This can happen if you’re on the free plan of your MCP client.
Some MCP clients and AI model providers limit how many tools a request can include. Connect with a toolset, such as ?toolsets=common or ?toolsets=readonly, so your client sees fewer Cortex tools.
Yes. The Cortex MCP can create and update many kinds of data, such as entities, Scorecards, Initiatives, and Workflows, if the user who owns your personal access token has the required permissions. It can’t delete data or run bulk operations. See Making changes with the Cortex MCP.
Ensure the value of your Cortex API token is valid in your configuration.
Ensure your OpenAPI spec path is correct when mounting.
Verify your Cortex API endpoint is accessible.
Ensure that you set up the remote MCP with a personal access token generated in Cortex. Remote MCP setup requires a personal access token, not an API key.
This is most likely caused by an issue within your network. It’s recommended to contact your organization’s infrastructure or security team to troubleshoot.
Last modified on November 5, 2026