What You Can Do With 6DuckLearn MCP: 26 Tools for AI Agents

Quick Answer

6DuckLearn exposes 26 MCP tools to an authenticated AI client. They cover connection health, tasks, saved research, Cornell Notes, data sources, market research, signal alerts, and Hong Kong credit-card comparison.

Twenty-two tools are read-oriented. Three tools propose protected writes. One continuation tool resumes the exact write only after the account owner approves it in 6DuckLearn Inbox.

This guide reflects the active public catalog on September 17, 2026. Your client only sees tools allowed by its current OAuth grant and the live server manifest.

Connect From Codex

Run the public setup helper:

npx github:kit18/6ducklearn-mcp setup-codex

The helper adds the hosted 6DuckLearn MCP server, then starts an OAuth login with the narrow scopes mcp:read, mcp:write, and approval:request.

Complete the browser sign-in and consent flow. Then ask Codex: “Check my 6DuckLearn MCP connection status.”

For another client that supports remote Streamable HTTP and OAuth, use https://6ducklearn.com/mcp. Follow that client’s connector setup flow and inspect the tools it receives.

Connection And Tasks

• get_mcp_connection_status checks the endpoint, authentication state, account identity status, scopes, and verification time without returning credentials.
• get_todo_list lists tasks owned by the connected account.
• create_todo proposes a new task. Account policy may require Inbox approval before the write can run.

Example prompt: “Check my connection, list my open tasks, and suggest the smallest next action. Do not create anything.”

Saved Knowledge And Data Sources

• search_pkm_articles searches saved articles or browses recent items.
• get_pkm_article reads one article selected from search results.
• list_cornell_notes lists Cornell Notes, with an optional topic filter.
• get_cornell_note reads one Cornell Note by its returned ID.
• build_with_ai_list_data_sources lists available 6DuckLearn Work data adapters.
• build_with_ai_fetch_data_sources fetches authorized source payloads through those adapters.

Example prompt: “Find my recent notes about onboarding, compare the evidence, and label anything stale or conflicting.”

Market Research And Hong Kong Cards

• search_assets finds stocks in the curated universe.
• get_quote returns the latest available daily quote snapshot.
• get_ohlcv returns cached price candles.
• get_market_news fetches recent market news through the current news adapter.
• screen_stocks applies bounded stock filters.
• get_industry_data returns sector or industry peer context.
• get_stock_important_dates reads cached earnings, dividend, split, or filing dates.
• search_saved_market_articles searches the connected user’s saved market research.
• get_watchlist lists the connected user’s watchlists and items.
• get_signal_lab_snapshot returns a read-only Signal Lab snapshot.
• get_symbol_research_context builds a bounded research bundle with freshness and data gaps.
• list_signal_alerts lists the connected user’s signal alerts.
• query_credit_card_catalog reads reviewed, published Hong Kong card evidence without ranking cards.
• recommend_credit_cards runs a deterministic comparison against that published catalog.
• create_signal_alert proposes a new watchlist-backed alert.
• update_signal_alert proposes a change to an existing alert.

Example prompt: “Build a dated research brief for 0700.HK. Show freshness, missing data, and a downside case. Do not give me a trade instruction.”

Example prompt: “Compare Hong Kong cards for my stated monthly spend. Use verified evidence only and show any missing fields.”

How Approval-Gated Writes Work

A protected write can return approval_required instead of changing data. 6DuckLearn stores the exact proposal and gives the client an approval ID plus an Inbox link.

The account owner reviews the locked proposal in Inbox. Chat confirmation alone is not approval, and the MCP connection does not request approval:decide.

After approval, resume_approved_write accepts only the server-issued approval ID. It cannot change the original tool arguments.

Approval records a human decision. The write is complete only when the resume result reports a successful terminal receipt.

What The Tools Do Not Do

Market tools support research and organization. They do not route orders, provide personalized investment advice, promise returns, or replace an independent review.

Hong Kong card results depend on the reviewed public catalog. Stale coverage or missing terms can produce a limited or no-result response instead of a forced winner.

Source adapters may return partial fields when a provider does not supply them. A missing value is different from a measured zero.

Tool availability can change by account, OAuth grant, policy, and server version. Treat the live client manifest as authoritative.

Troubleshooting

If no tools appear, reconnect OAuth and refresh or restart the client. Confirm that the configured URL is https://6ducklearn.com/mcp.

If a tool is missing, do not invent a replacement. Ask the client to list its active MCP tools and confirm the required scope is present.

If a write is pending, decide it in 6DuckLearn Inbox. Resume only with the returned approval ID, and do not replay a write whose outcome is unknown.

Proof Level And Limitations

• Proof level: Public product guide checked against the active 26-tool hosted manifest and public Codex setup flow on September 17, 2026.
• Demonstrates: Available tool names, public connection steps, OAuth boundaries, and the approval-resume model.
• Does not claim: Customer adoption, productivity gains, investment performance, complete provider coverage, or guaranteed results.
• Public source: 6DuckLearn MCP connector on GitHub.

Start With A Read

Connect the MCP, check its status, and begin with a read-only prompt. Add a protected write only when you understand the exact proposal and are ready to review it in Inbox.