MCP Connector
Telumin doesn’t sell a REST API. It ships one connector, built on the Model Context Protocol — the open standard AI agents already speak. Connect it once and everything Telumin knows about your book becomes tools your agent can call: in claude.ai, in Claude Code, or in an agent you built yourself.
One URL, OAuth sign-in, no keys
There is nothing to provision. The connector authenticates with OAuth 2.1 against your Telumin account — your agent registers itself, opens the consent screen, and you approve the scopes. No API keys to create, copy, rotate, or leak.
Transport
Streamable HTTP (single POST endpoint)
Auth
OAuth 2.1 — authorization code + PKCE (S256)
Client setup
Dynamic client registration (RFC 7591) — automatic
Tokens
1-hour access tokens, rotating 30-day refresh
claude.ai & Claude Desktop
In claude.ai go to Settings → Connectors → Add custom connector, paste the URL above, and sign in with your Telumin login when the consent screen opens. From then on Claude can answer with your real holdings, real threats, and the methodology behind them.
Things to ask once connected
“Any threats on my holdings since yesterday?”
“What’s sitting in my action queue — and which would you reject?”
“Run the Telumin methodology on NVDA and save the analysis.”
Claude Code
Claude Code speaks MCP natively — one command adds Telumin to every session:
The browser opens Telumin’s consent screen; approve it and the tools are live. Your terminal becomes the same protect loop you get in the app — positions, threats, the action queue, deep company research — woven into whatever else you’re doing there.
Works well in Claude Code
“Check my Telumin action queue and summarize what needs me today.”
“Pull the subgraph for AAPL, apply the methodology, save the analysis.”
“Add the tickers from this CSV to my Telumin watchlist.”
Bring your own agent
Any MCP client that supports Streamable HTTP and OAuth works — the Claude Agent SDK, the official MCP SDKs, or your own client. The handshake is fully self-describing, so a compliant client needs only the URL:
Scopes are granted on the consent screen and enforced per tool. Request only what your agent needs:
| Scope | Grants |
|---|---|
| portfolio:read | Positions and signals |
| threats:read | Threats matched to your book |
| actions:read / actions:write | Read the action queue / decide items |
| watchlist:write | Add or remove watched tickers |
| investigations:trigger | Queue engine investigations |
| research:read / research:write | Knowledge graph / save analyses |
The tool surface
Every tool is isolated to the signed-in account — an agent holding your token sees your book and nobody else’s.
Portfolio
| get_positions | Your positions — symbol, qty, cost, market value, P&L. | portfolio:read |
| get_signals | Recent intelligence signals: headline, direction, tickers. | portfolio:read |
| update_watchlist | Add or remove a ticker on your watchlist. | watchlist:write |
Threats & actions
| list_portfolio_threats | Threats matched to your holdings and watchlist, newest first. | threats:read |
| get_action_queue | Pending proposed actions — the engine proposes, you decide. | actions:read |
| acknowledge_or_snooze_action | Accept, reject, or snooze one queue item. | actions:write |
| trigger_investigation | Queue the engine to investigate an entity on its next cycle. | investigations:trigger |
Research & analysis
| get_company_subgraph | A company’s knowledge-graph neighborhood: co-exposed names, shared threats, your exposure overlay. | research:read |
| get_analysis_methodology | The structured 6-dimension value methodology the agent should follow. | any signed-in token |
| save_analysis | Save a finished analysis back to Telumin — rendered, evidence-stamped, shareable by link. | research:write |
Access & plans
Connecting is free after a workbook import — methodology and research tools work on any signed-in account. Tools that read your live book (positions, threats, the action queue) need an active Telumin Plus subscription. When a tool is gated it never hard-errors: it replies with a plain message and a subscribe link, so agent conversations degrade gracefully instead of crashing.
Two minutes to connect
Import your workbook free, then add the connector in claude.ai or run one claude mcp add in your terminal.