An MCP server for the CDTM Community Portal, so you can search CDTM students and alumni from Claude Code, Claude Desktop, or any other MCP client.
Instead of clicking through filter dropdowns, you just ask:
"find me CDTM people building AI agents who can help with fundraising"
and Claude figures out the right tags, queries the portal, and gives you names, headlines, and profile links.
Hi, I'm Hamza (Fall 2017, co-founder at ZenML). The community portal is genuinely useful — a few thousand people with skills, interests, and "can support with" tags — but I never remember to open it. I do live in Claude Code all day, so I wired the portal into it as an MCP server. Now finding "alumni in Munich who know venture capital and offer mentoring" is one sentence away, and Claude can cross-reference results with whatever else I'm working on.
If you find this useful, give it a star ⭐ — if enough people do, I'll ask the CDTM team for a proper way to integrate this into the internal app (API tokens instead of the cookie dance below).
You need Python 3.10+ and uv (or plain pip).
git clone https://github.com/htahir1/cdtm-mcp.git
cd cdtm-mcpThe portal sits behind oauth2-proxy, and there's no public API or token system (yet — see the star pitch above). So the server authenticates with your own browser session cookie:
- Log in to internal.cdtm.com in your browser.
- Open DevTools (F12) → Network tab → reload the page.
- Click any request to internal.cdtm.com → Headers → find the
Cookierequest header. - Copy the part that starts with
_HOST-oauth2-proxy=(it's long). - Create a
.envfile in the repo root:
CDTM_COOKIE='_HOST-oauth2-proxy=<paste the value here>'
The cookie expires after a while. When the tools start failing with an auth error, repeat the steps. .env is gitignored — keep it that way, this cookie is your login.
claude mcp add --scope user cdtm -- uv run --directory /path/to/cdtm-mcp python server.pyThat's it. Open a new Claude Code session and ask it something.
For Claude Desktop, add the equivalent command to your claude_desktop_config.json.
| Tool | What it does |
|---|---|
search_members |
Filter by name, skills, interests, "can support with", "looking for", companies, education, class, location. Paginated. |
list_tags |
Browse the portal's tag taxonomy (110 skills, 134 interests, ~3000 companies, ...) |
get_member_profile |
Full profile for one person: bio, tags, contact email |
refresh_taxonomy |
Re-fetch the tag list if it looks stale |
Example prompts that work well:
- "find CDTM people into agentic AI"
- "who from the Fall 2017 class is in Berlin?"
- "alumni who can support with pitch feedback and know climate tech"
- "pull the full profile for the person I just found"
Nothing fancy. The portal is a Next.js app whose search runs through its own JSON endpoint (/api/protected/directory?skills=414&interests=62&q=...). This server:
- pulls the tag taxonomy (names → IDs) out of the
/homepage payload once per session - resolves the human-readable tag names Claude picks into IDs
- calls the directory endpoint and returns clean JSON
- parses profile pages server-side for
get_member_profile
No scraping of anything you couldn't see in your own browser, no data stored anywhere.
This reads member-provided data through your own authenticated session. It's for finding people to talk to — not for bulk-exporting the directory or spamming anyone. Don't be that person.
This is an unofficial community project and isn't affiliated with the CDTM team. If they have any concerns about it, or want to talk about a proper API, I'm happy to hear from them.