MCP server

Stateless Streamable HTTP (JSON-RPC 2.0, protocol 2025-06-18 with fallbacks). One URL, https://uplift.page/mcp, for everyone: browsing, evaluating, and installing full skills work anonymously. The first time your agent calls a workspace tool, the server asks it to sign you in: you enter your email code, approve the agent once, and every call then runs as you against the workspaces you belong to. A workspace key (upk_…) does the same for automation and CI.

Connect your agent

Every client needs only the URL. When a tool needs your workspace, the server answers 401 and your client opens uplift.page to sign in with your emailed code; you allow the app once, and it never sees anything outside the workspaces you belong to. No client ID, secret or key to paste. Revoke an app later under Account → Connected agents.

Name: uplift.page
URL: https://uplift.page/mcp
(leave Advanced settings empty)

claude mcp add -s user --transport http uplift https://uplift.page/mcp

MCP server URL: https://uplift.page/mcp
Authentication: OAuth

[mcp_servers.uplift]
url = "https://uplift.page/mcp"

{
  "mcpServers": {
    "uplift": { "url": "https://uplift.page/mcp" }
  }
}

{
  "servers": {
    "uplift": { "type": "http", "url": "https://uplift.page/mcp" }
  }
}

{
  "mcpServers": {
    "uplift": { "serverUrl": "https://uplift.page/mcp" }
  }
}

{
  "mcpServers": {
    "uplift": { "httpUrl": "https://uplift.page/mcp" }
  }
}

https://uplift.page/mcp

{
  "context_servers": {
    "uplift": { "url": "https://uplift.page/mcp" }
  }
}

npx @modelcontextprotocol/inspector

Other clients: add a remote (Streamable HTTP) MCP server with the URL above. Any client that supports MCP authorization signs in the same way; field names vary, so check your client's MCP settings if one differs.

Automation, CI and hooks

Machines that can't sign in in a browser use a workspace key from dashboard.uplift.page → Telemetry keys, as a literal Authorization header in the client's own config. Desktop apps and scheduled runs don't read your shell profile, so an environment variable there arrives empty. A key in the header also turns off the sign-in flow in most clients.

claude mcp add -s user --transport http uplift https://uplift.page/mcp --header "Authorization: Bearer upk_YOUR_KEY"

{
  "mcpServers": {
    "uplift": {
      "type": "http",
      "url": "https://uplift.page/mcp",
      "headers": { "Authorization": "Bearer upk_YOUR_KEY" }
    }
  }
}

curl -s https://uplift.page/mcp -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"pull_skill","arguments":{"slug":"code-style"}}}' \
  | jq -r '.result.structuredContent.markdown' > skills/code-style/SKILL.md

curl -s -X POST https://uplift.page/mcp -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jq '.result.tools[].name'

Public tools

Skill-library tools

List, install, and score skills: public skills need no account; signing in (or a workspace key) unlocks your workspace's private side. People in several workspaces pass workspace to pick one of theirs. Writes need a paid plan.

A typical agent flow: recommend_stack with the build intent → adopt the mini prompt → ship → send_feedback with the outcome. For skills: list_skills → pull_skill → write the markdown to skills/<slug>/SKILL.md. The instructions field returned by initialize tells models exactly this.