MCP Server

Connect GitHub Copilot, Claude Code, and other AI assistants directly to your product context

Delvyn Studio MCP Server
New

The Delvyn Studio MCP (Model Context Protocol) server exposes your product data directly to AI coding assistants. Connect GitHub Copilot, Claude Code, Cursor, or any MCP-compatible tool to give your AI assistant instant access to product vision, OKRs, agent specs, discovery insights, and more — without copying and pasting context manually.

What You Get

  • • Live product vision and strategy context
  • • Current OKR objectives and key result progress
  • • Full agent specifications with acceptance criteria
  • • Discovery insights and learning captures

Supported Clients

  • • Claude Desktop
  • • GitHub Copilot (VS Code agent mode)
  • • Claude Code (terminal)
  • • Cursor IDE
  • • Any MCP-compatible AI assistant
Prerequisites

Before You Begin

  • • A Delvyn Studio account (any plan)
  • • An MCP API key — generate one in Settings → API Access
  • • An MCP-compatible AI client:
    • — Claude Desktop (requires Node.js 18+)
    • — VS Code with the GitHub Copilot extension (Chat agent mode)
    • — Claude Code CLI
    • — Cursor IDE
Step 1: Generate an API Key
  1. 1

    Open API Access settings

    Navigate to Settings → API Access

  2. 2

    Generate a new key

    Click "Generate API Key" and copy the key — it starts with dvn_

  3. 3

    Copy the generated MCP config

    The API Access page shows a ready-to-use MCP config snippet. Copy it for the next step.

Key Security

Store your API key in your MCP config file or a secrets manager. Do not commit it to source control. You can revoke and regenerate keys at any time from the API Access settings.

Step 2: Configure GitHub Copilot (VS Code)

VS Code stores MCP server configuration in .vscode/mcp.json inside your project, or in your user settings. The project-level file is recommended and can be shared with your team (redact the API key before committing).

Create or update .vscode/mcp.json

{
  "servers": {
    "delvyn": {
      "type": "http",
      "url": "https://api.delvynstudio.com/mcp",
      "headers": {
        "Authorization": "Bearer dvn_YOUR_API_KEY_HERE"
      }
    }
  }
}

Replace dvn_YOUR_API_KEY_HERE with your actual API key. The exact snippet (with your key pre-filled) is available in Settings → API Access.

Enable Agent Mode in VS Code

  1. 1

    Open Copilot Chat

    Press Ctrl+Alt+I (or ⌘⌥I on Mac) to open GitHub Copilot Chat

  2. 2

    Switch to Agent Mode

    Click the mode selector in the chat panel and choose Agent

  3. 3

    Verify the server appears

    Click the tools icon () in the chat panel. You should see delvyn listed as an available MCP server with its tools.

  4. 4

    Ask Copilot about your product

    Try: "What are the current OKRs for our product?" or "Show me the agent spec for the search feature"

Step 3: Configure Claude Code

Claude Code reads MCP server configuration from .claude/mcp.json in your project root, or from the global config at ~/.claude/mcp.json.

Add via CLI

claude mcp add delvyn \
  --transport http \
  --url https://api.delvynstudio.com/mcp \
  --header "Authorization: Bearer dvn_YOUR_API_KEY_HERE"

Or edit the config manually

Create or update .claude/mcp.json:

{
  "mcpServers": {
    "delvyn": {
      "type": "http",
      "url": "https://api.delvynstudio.com/mcp",
      "headers": {
        "Authorization": "Bearer dvn_YOUR_API_KEY_HERE"
      }
    }
  }
}

Verify the connection

# Start Claude Code in your project folder
claude

# In the Claude session, verify tools are available
/mcp

You should see delvyn listed with 17 available tools.

Step 4: Configure Claude Desktop

Why mcp-remote?

Claude Desktop only supports local stdio-based MCP servers in its configuration file. The mcp-remote package (open source, 200k+ weekly npm downloads) acts as a bridge — it runs locally as a stdio server and forwards requests to the remote Delvyn Studio MCP endpoint. Requires Node.js 18+.

Edit the config file at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows).

macOS / Linux

{
  "mcpServers": {
    "delvyn": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://api.delvynstudio.com/mcp",
        "--header",
        "Authorization: Bearer dvn_YOUR_API_KEY_HERE",
        "--transport", "http-only"
      ]
    }
  }
}

Windows

{
  "mcpServers": {
    "delvyn": {
      "command": "cmd",
      "args": [
        "/c", "npx", "-y", "mcp-remote",
        "https://api.delvynstudio.com/mcp",
        "--header",
        "Authorization: Bearer dvn_YOUR_API_KEY_HERE",
        "--transport", "http-only"
      ]
    }
  }
}

On Windows, cmd /c is required because Claude Desktop does not invoke npx correctly otherwise.

Verify the connection

  1. 1Restart Claude Desktop completely (quit from system tray, then relaunch)
  2. 2Look for the hammer icon in the bottom-right of the message input box
  3. 3Click the hammer icon — you should see delvyn listed with its tools
  4. 4Try: “What are the current OKRs for our product?”
Available Tools (17)

Product Context
3 tools

get_product_context

Retrieve full context for a product including vision, strategy, and guardrails

get_product_vision

Get the vision statement and strategic direction for a product

get_product_strategy

Retrieve the strategic initiatives and positioning for a product

OKRs
4 tools

get_current_objectives

List all active objectives for the current OKR cycle

search_objectives

Search objectives by keyword or product

get_key_result_progress

Get the current progress, confidence, and status for a key result

list_cycles

List all OKR cycles for the organization

Agent Specs
2 tools

get_agent_spec

Retrieve a complete agent specification with acceptance criteria and context

search_agent_specs

Search agent specs by keyword, product, or status

Discovery
4 tools

search_ideas

Search product ideas by keyword or status

get_idea_details

Get full details for a product idea including notes and linked specs

search_discoveries

Search discovery projects by keyword or status

get_discovery_details

Get full details for a discovery project including findings

Learning Captures
3 tools

log_learning

Log a new learning or insight to a discovery project

get_learning_captures

Retrieve learning captures for a discovery project

get_spec_quality_metrics

Get quality metrics for agent specs in a product

Teams
1 tool

list_teams

List all teams in the organization

Usage Examples

Get product context before implementing

"Before you start coding, check our product vision and current OKRs for the Search product. Then look up the agent spec for the 'Advanced filtering' feature and implement it according to the acceptance criteria."

Log discoveries while coding

"I found that the pagination approach in the current codebase conflicts with the strategy guardrails. Log this as a learning capture in the 'Search UX' discovery project."

Check OKR progress

"What's the current progress on the Q2 OKRs? Which key results are behind and what agent specs are linked to them?"

Context-aware code review

"Review this PR. First check the agent spec it's linked to, then verify the implementation matches the acceptance criteria and doesn't violate any product strategy guardrails."
Security Best Practices

Keep your API key out of source control

Add .vscode/mcp.json to .gitignore, or use environment variable substitution if your MCP client supports it.

Create one key per device or client

Use separate API keys for your laptop, CI, and each team member. This makes it easy to revoke access for a single device without affecting others.

Rotate keys regularly

Revoke and regenerate keys every few months, or immediately if you suspect a key has been exposed.

MCP server is read-only by default

Most tools only read data. The log_learning tool writes a learning capture. No tools can modify your products, OKRs, or agent specs.

Troubleshooting

Server not appearing in VS Code

Ensure you are in Agent mode (not Chat or Edits). Reload the VS Code window after creating or modifying .vscode/mcp.json.

Authentication error (401)

Your API key may be expired, revoked, or incorrectly formatted. Generate a new key from Settings → API Access and update your config.

Tools return empty results

Make sure you have products, OKRs, or agent specs created in Delvyn Studio. The MCP server returns data for your account's organization only.

Claude Code can't find the server

Run claude mcp list to verify the server is registered. If not, re-run the claude mcp add command above.

Claude Desktop shows “could not be loaded”

Verify Node.js 18+ is installed (node --version). Delete the ~/.mcp-auth folder to clear cached state. Add "--debug" to the args array for detailed logs in ~/.mcp-auth/. On Windows, check logs at %LOCALAPPDATA%\Claude\Logs\mcp.log.