Guide

How to Connect Claude Code to DAVE over MCP

Summary

Claude Code connects to your DAVE workspace over a hosted HTTPS MCP server using a per-user key. Nothing is installed locally. This guide walks through generating your MCP key, choosing a capability preset, and adding the server to Claude Code so your AI coding assistant can interact with your workflows, agents, and tasks.

The steps

  1. Navigate to MCP key management

    MCP keys in DAVE are per-user credentials. Each user generates their own key, scoped to their account and the capability preset they choose. An Admin can also generate and manage MCP keys on behalf of other users. The key management interface is inside the API Management section of the Admin panel.

    Do this
    • In the sidebar, click Admin.
    • Click API Management.
    • Locate the MCP Keys tab or section within API Management.
    • Confirm you can see a list of existing MCP keys (or an empty state if none have been created yet).
    Example

    If the Admin option is not visible in your sidebar, your account does not have the Admin role. For non-Admin users who need to generate their own MCP key, check whether a self-service key generation option is available under your account settings. If not, ask an Admin to generate a key on your behalf.

    Best practice

    Generate one MCP key per person and per AI assistant. Do not share MCP keys between users or between Claude Code and other AI assistants such as Claude Desktop or VS Code. Per-user keys mean the audit log records which person's assistant performed each action, not just that an MCP action occurred. Shared keys make the audit trail ambiguous.

  2. Choose a capability preset and generate the key

    DAVE's MCP integration exposes 34 tools across four capability presets. When you generate an MCP key, you select a preset that determines which tools the key grants access to. The preset is set at key creation and cannot be changed afterward: to change the preset, generate a new key.

    The four presets follow a least-privilege model. Each preset grants access to a defined subset of the 34 available tools, matched to a specific use case. Selecting the narrowest preset that covers your intended use is the recommended approach: it limits what Claude Code can do through MCP to exactly what you need, and nothing more.

    The preset options and the tool categories they cover are sourced from the DAVE MCP documentation. The specific preset names and their exact tool lists are not detailed in the feature summary document: see the dedicated MCP help documentation for the full preset-to-tool mapping. The key facts are: four presets exist, each grants a subset of the 34 total tools, and the preset is locked at key creation.

    After selecting a preset, click the button to generate the key. DAVE displays the key value once, in full, immediately after generation. Copy it now: it will not be shown again. Store it in a secure location such as a password manager before closing the dialog.

    Do this
    • Click Generate New MCP Key (or the equivalent button in the MCP Keys section).
    • Review the four capability preset options and select the one that matches your intended use for Claude Code.
    • Optionally, give the key a descriptive name (for example: "Claude Code - [your name]") so it is identifiable in the audit log and key list.
    • Click Generate (or Create).
    • When the key value is displayed, copy it immediately to a secure location. It will not be shown again.
    • Close the dialog. The key now appears in the MCP Keys list with its preset, creation date, and last-used timestamp.
    Example

    A developer who wants Claude Code to be able to read workflow definitions and run instances but not modify agent configurations would select the preset that covers workflow and instance tools, rather than the full-access preset. This limits the blast radius if the key is ever compromised: Claude Code can only do what the preset allows.

    Best practice

    Name every MCP key clearly at creation. A key named "Claude Code - Jane Smith" is immediately identifiable in the audit log and key list. A key named "key-1" is not. If you need to revoke a key later, a clear name makes it easy to find the right one without guessing. Revocation takes effect within minutes, even on open sessions.

  3. Find your DAVE MCP server URL

    DAVE hosts the MCP server over HTTPS. There is nothing to install locally: Claude Code connects to DAVE's hosted endpoint over the internet using your MCP key for authentication. The MCP server URL is specific to your DAVE workspace (tenant).

    The exact format of the MCP server URL and where it is displayed in the platform are not detailed in the feature summary document. The URL is typically displayed alongside the generated key, or in the MCP Keys section of API Management. Look for a field labelled MCP Server URL, Endpoint, or Server in the key detail view.

    The URL will be an HTTPS address specific to your tenant. Copy it exactly as shown: it is case-sensitive and must be used without modification in the Claude Code configuration step that follows.

    Do this
    • In the MCP Keys section of API Management, locate the MCP Server URL or equivalent field.
    • Copy the full HTTPS URL. It is specific to your DAVE workspace.
    • Keep both the MCP key value (copied in Step 2) and the server URL ready for the next step.
    Example

    The MCP server URL will look something like https://your-workspace.hellodave.ai/mcp or a similar HTTPS address. The exact format depends on your tenant configuration. Use the URL shown in the platform, not a guessed or constructed one.

    Best practice

    Do not construct the MCP server URL manually. Use the URL shown in the platform. If the URL is not visible in the MCP Keys section, check the MCP documentation or ask your DAVE Admin. An incorrect URL will cause Claude Code to fail to connect, with a connection refused or not found error.

  4. Add the MCP server to Claude Code

    Claude Code supports MCP servers through its configuration file or through the claude mcp add command. The connection requires two values: the MCP server URL (from Step 3) and your MCP key (from Step 2). Claude Code sends the key as an authorization header on every MCP request.

    Claude Code's MCP configuration accepts servers in the following shape, added to the mcpServers object in your Claude Code configuration:

    {
     "mcpServers": {
     "dave": {
     "type": "http",
     "url": "YOUR_DAVE_MCP_SERVER_URL",
     "headers": {
     "Authorization": "Bearer YOUR_MCP_KEY"
     }
     }
     }
    }

    Replace YOUR_DAVE_MCP_SERVER_URL with the URL copied in Step 3, and YOUR_MCP_KEY with the key value copied in Step 2.

    Alternatively, if Claude Code supports the claude mcp add CLI command in your version, you can add the server from the terminal. Consult the Claude Code documentation for the exact CLI syntax for HTTP MCP servers with authorization headers, as the syntax varies by version.

    Do this
    • Open your Claude Code configuration file (typically ~/.claude/claude_desktop_config.json or the equivalent for your installation).
    • Add a new entry to the mcpServers object using the shape shown above, substituting your DAVE MCP server URL and key.
    • Save the configuration file.
    • Restart Claude Code (or reload the MCP server list if Claude Code supports hot-reload).
    • Confirm the DAVE MCP server appears in Claude Code's connected servers list.
    Example

    After adding the configuration and restarting Claude Code, type a prompt such as: "List the active workflows in my DAVE workspace." If the connection is working, Claude Code will use the DAVE MCP tools to fetch and return the list. If it returns an error or says it cannot access DAVE, check that the URL and key are correct and that the key has not been revoked.

    Best practice

    Store the MCP key in the configuration file using your operating system's secret management where possible (for example, macOS Keychain or a secrets manager), rather than as a plaintext string in the config file. The key grants access to your DAVE workspace: treat it with the same care as a password. If you suspect the key has been exposed, revoke it immediately in DAVE's MCP Keys section and generate a new one.

  5. Verify the connection and confirm audit logging

    After adding the MCP server to Claude Code and restarting, verify that the connection is working and that Claude Code's actions are appearing in the DAVE audit log. Both checks together confirm that the integration is working correctly and that governance is active.

    The audit log records every MCP action with the same context as any other user action: the acting key, the tool called, the parameters, and the timestamp. Because MCP keys are per-user, the audit log entry identifies which user's Claude Code instance performed the action. This is the governance guarantee: Claude Code works inside your DAVE audit boundary, not outside it.

    Do this
    • In Claude Code, send a simple prompt that uses a DAVE MCP tool, such as asking Claude Code to list your DAVE workflows or describe a specific agent.
    • Confirm Claude Code returns a response that reflects real data from your DAVE workspace.
    • In DAVE, navigate to Admin, then Settings and click the Audit Log tab.
    • Look for a recent event with an event type related to the MCP action you just performed (for example, a workflow list read or an agent read).
    • Expand the event row and confirm the Actor field shows your MCP key identifier.
    • Confirm the event timestamp matches the time of your Claude Code prompt.
    Example

    A developer asks Claude Code: "What workflows are active in my DAVE workspace?" Claude Code calls the DAVE MCP tool for listing workflows and returns the results. The developer then checks the DAVE Audit Log and finds a read event timestamped to within seconds of the prompt, with the MCP key as the actor. The integration is confirmed working and audit-logged.

    Best practice

    Run this verification step immediately after connecting Claude Code, not days later. If the audit log shows no MCP events after a successful Claude Code response, check whether your tenant's compliance standards are configured to log read-access events (SOC 2 or GDPR mode in Admin, then Settings, then System Parameters). Without those standards enabled, read-only MCP actions may not generate audit events. Mutating actions (creating or updating records) are always logged regardless of compliance mode.

What MCP gives Claude Code access to

DAVE's MCP integration exposes 34 tools to connected AI assistants. These tools cover the core operational surface of the platform: reading and managing workflows, working with agents, interacting with tasks, and accessing workspace data. When Claude Code is connected over MCP, it can use these tools in response to natural-language prompts, without you switching between the Claude Code terminal and the DAVE browser interface.

The 34 tools are organized into four capability presets. Each preset grants access to a defined subset of tools, matched to a specific use case. The preset you select when generating your MCP key determines what Claude Code can do in your workspace. Selecting a narrower preset limits Claude Code's access to exactly what you need: this is the least-privilege principle applied to AI assistant access.

Four AI assistants are supported for MCP connection: Claude Code, Claude Desktop, VS Code, and JetBrains. This guide covers Claude Code specifically. The connection mechanism (hosted HTTPS, per-user key, no local installation) is the same for all four. If you use multiple AI assistants, generate a separate MCP key for each one so the audit log can distinguish their actions.

Nothing is installed locally to enable the MCP connection. DAVE hosts the MCP server over HTTPS. Claude Code connects to it over the internet using your key. Your local development environment does not need any DAVE-specific software, daemon, or proxy. The only local change is adding the server configuration to Claude Code's MCP settings.

The security model: keys, presets, and the audit trail

DAVE's MCP security model has three layers that work together to give Claude Code useful access while keeping your workspace under governance control.

Per-user keys. Each MCP key is tied to a specific user account. When Claude Code uses an MCP tool, the action is recorded in the audit log against that key and the user it belongs to. This means the audit trail distinguishes between actions taken by different people's AI assistants, not just between human and AI actions. If two developers both use Claude Code connected to the same DAVE workspace, their actions are separately attributable in the audit log.

Capability presets. The preset selected at key creation determines which of the 34 tools the key can access. A key with a read-only preset cannot create or modify records, even if Claude Code sends a request to do so: the request is rejected at the API level. This is enforced on every request, not just at connection time. Revocation also takes effect within minutes on open sessions: if you revoke an MCP key, Claude Code loses access quickly, without needing to restart.

Full audit logging. Every action Claude Code takes through MCP is recorded in the DAVE audit log with context: the tool called, the parameters, the response status, the timestamp, and the acting key. This is the same audit infrastructure that logs human user actions. AI assistant actions are not a separate, less-audited category: they are first-class audit events. For tenants with SOC 2 or GDPR compliance modes enabled, MCP read-access events are also logged, giving a complete picture of what Claude Code accessed, not just what it changed.

The combination of per-user keys, capability presets, and full audit logging means Claude Code works inside your DAVE governance boundary. It is not a side channel that bypasses your access controls: it is a governed integration that respects them.

What MCP is not: common misconceptions

Before connecting Claude Code over MCP, it is worth being clear about what the integration does not do, to avoid misconfigured expectations.

MCP is not a way to run workflows from Claude Code. The MCP tools give Claude Code access to workflow definitions, agent configurations, and workspace data. Whether Claude Code can trigger workflow instances depends on which capability preset the key was generated with. Even with a preset that includes instance creation, Claude Code is creating an instance in DAVE: the workflow runs in DAVE, not in Claude Code. Claude Code is the interface, not the execution engine.

MCP is not self-hosting. Connecting Claude Code to DAVE over MCP does not give you a self-hosted DAVE instance. DAVE is fully managed and hosted by voolama LLC. The MCP connection is a remote API connection to the hosted platform. Bringing your own Claude Code installation is not the same as self-hosting DAVE.

MCP keys are not the same as API keys. DAVE has a separate API key system for programmatic access to its REST API. MCP keys are specifically for MCP tool access by AI assistants. They use different authentication mechanisms and grant access to different surfaces. Do not use an MCP key where an API key is expected, or vice versa.

The MCP connection does not give Claude Code access to your AI provider keys. DAVE stores AI provider credentials (OpenAI keys, Anthropic keys, and so on) in an encrypted vault using AES-256-GCM authenticated encryption. These credentials are never returned through the API and are not accessible through MCP tools. Claude Code can instruct DAVE to run an agent, but it cannot read the API key that agent uses.

Frequently asked questions

Do I need to install anything locally to connect Claude Code over MCP?
No. DAVE hosts the MCP server over HTTPS. The only local change is adding the server URL and key to Claude Code's MCP configuration. No daemon, proxy, or DAVE-specific software is installed on your machine.

Can I use the same MCP key for Claude Code and Claude Desktop?
You can, but you should not. Per-user, per-assistant keys mean the audit log can distinguish which assistant performed each action. A shared key makes the audit trail ambiguous. Generate a separate key for each AI assistant you connect.

What happens if I select the wrong capability preset?
The preset is locked at key creation. To change it, revoke the existing key and generate a new one with the correct preset. Revocation takes effect within minutes.

Are Claude Code's MCP actions logged in the DAVE audit trail?
Yes. Every MCP action is recorded in the audit log with the acting key, the tool called, the parameters, and the timestamp. For read-access events to be logged, your tenant needs SOC 2 or GDPR compliance mode enabled in Admin, then Settings, then System Parameters. Mutating actions are always logged.

Can Claude Code access my AI provider API keys through MCP?
No. Provider credentials are stored in an encrypted vault and are never returned through the API or through MCP tools. Claude Code can ask DAVE to run an agent, but it cannot read the credentials that agent uses.

What should I do if Claude Code cannot connect to the DAVE MCP server?
Check three things in order: that the MCP server URL is correct and copied exactly from the platform; that the MCP key value is correct and has not been revoked; and that Claude Code's MCP configuration file has been saved and Claude Code has been restarted. If all three are correct and the connection still fails, check whether your network allows outbound HTTPS connections to the DAVE server URL.

How do I revoke an MCP key?
Navigate to Admin, then API Management, then the MCP Keys section. Find the key in the list and use the revoke or delete action. Revocation takes effect within minutes, even on open Claude Code sessions.

How to Connect Claude Code to DAVE over MCP