The manager is the reason to look at this: isolated environments per server plus one command that writes config for six different clients is the part everybody rebuilds badly. Of the bundled servers, jira-helper is the substantial one, and loadbearing-youtube is the unusual one — it looks for the load-bearing claims in a video rather than summarising it.
A monorepo of Python MCP servers with an installer, `mcp-manager`, that gives each server its own isolated uv-managed environment and writes the client configuration for you. `mcp-manager sync` covers Cline, Claude Desktop, Claude Code, VS Code native MCP, Codex and Antigravity, and leaves hand-written entries in those config files alone.
- Jira-helper — the largest of the bundled servers, covering issues (`get_issue_details`, `create_jira_ticket`, `update_jira_issue`, `transition_jira_issue`, `change_issue_assignee`), search (`search_jira_issues`, `validate_jql_query`), links (`create_issue_link`, `create_epic_story_link`), worklogs and estimates (`log_work`, `get_work_logs`, `get_time_tracking_info`), attachments (`upload_file_to_jira`, `list_issue_attachments`) and Confluence pages (`list_confluence_spaces`, `get_confluence_page`, `search_confluence_pages`, `create_confluence_page`, `update_confluence_page`), across multiple Jira instances
- Worldcontext — the facts a model does not have: `get_current_datetime`, `get_stock_quote`, `get_stock_market_overview`, `get_news_headlines`, `get_latest_tool_versions`, `get_python_package_version` and a combined `get_context_summary`
- Loadbearing-youtube — `analyze_video` extracts a transcript and surfaces the claims, decisions, tradeoffs and recommendations a video's conclusion actually rests on, with timestamps; long videos run as a background job you collect with `get_analysis_result`, and `get_video_transcript` returns the raw text
- Mcpservercreator — `create_mcp_server` scaffolds a new server from a code snippet, with `list_installed_servers` to see what is already there
- A `template` server to fork when you are writing your own, and a `mcp-manager sync --platform` flag when you only want to update one client
Per-server credentials, kept in `~/.config/mcp-manager/servers/<name>/config.yaml` — jira-helper needs an Atlassian API token. Install `uv`, then `uv tool install "git+https://github.com/dawsonlp/BaseMcpServer.git#subdirectory=utils/mcp_manager"`, then `mcp-manager install <name> --source ./servers/<name>` for each server you want. macOS and Linux are supported and tested; Windows is explicitly out of scope. Every server registers plain tool functions through the MCP Python SDK, so the template is a small file rather than a framework.
One command — uv tool install "git+https://github.com/dawsonlp/BaseMcpServer.git#subdirectory=utils/mcp_manager"