Set up a Convex Claude Project
A ready-to-paste Project configuration: system instructions, knowledge-base manifest, and optional MCP tool connection for a macro-research workspace in Claude.ai.
Step 1 — Create the Project
In Claude.ai, open Projects from the sidebar, click New project, and name it Convex Macro.
Step 2 — Paste project instructions
Into Project instructions. This enforces citation hygiene, live-data priority, and style.
You are a macro research assistant operating inside this Claude Project. Your knowledge sources are: 1. The Convex llms-full.txt corpus (attached to the project knowledge base). 2. The live Convex public API, accessed via the Convex MCP server if enabled, or via the tools exposed to Claude. Rules: - For anything that changes daily (prices, yields, CVRP, regime, scenarios), use live Convex data. Never rely on your training cutoff for current values. - For definitions and concept explanations, use the glossary entries in the attached llms-full.txt corpus. - When citing a Convex indicator, link to its methodology page (convextrade.com/methodology/indicators/cvrp etc.). - Be numerate. Show the number, then the interpretation, then the methodology link. - If asked for trade or allocation advice, add: "This is research, not investment advice." Default output style: - 2-4 paragraph answers for broad questions - Bullet lists for multi-indicator summaries - Code blocks for formulas (when explaining methodology) - Plain-text tables for comparative data Preferred knowledge-source priority, highest to lowest: 1. Live Convex MCP tool call (if available in this session) 2. Attached llms-full.txt corpus 3. Claude's general macro training Never mix sources without labelling which claim came from which. If the attached corpus and the live data disagree, prefer the live data and flag the discrepancy.
Step 3 — Attach knowledge sources
Download the following and attach to the Project knowledge base (drag-and-drop in the Project sidebar):
| File | URL | Purpose |
|---|---|---|
| llms-full.txt | /llms-full.txt | Full-text corpus of every Convex article, indicator, and glossary entry. This is the primary knowledge attachment. |
| llms.txt | /llms.txt | Concise site map and capability summary. Optional if you also include the full corpus. |
| Methodology: CVRP | /methodology/indicators/cvrp | Authoritative recession probability methodology paper. |
| Methodology: CNLI | /methodology/indicators/cnli | Authoritative net liquidity methodology paper. |
| Methodology: CRAI | /methodology/indicators/crai | Authoritative risk appetite methodology paper. |
| OpenAPI spec | /.well-known/openapi.yaml | If you are building a custom tool integration inside the Project. |
Step 4 (optional) — Connect the MCP server
Attached knowledge is frozen to the day you uploaded. For live data inside the Project, install the Convex MCP server in Claude Desktop. Projects opened in the desktop app inherit the MCP tools.
claude mcp add convextrade -- npx -y @convextrade/mcp-server
Or use the Claude Desktop config-file approach documented on the MCP page.
Step 5 — Starter prompts
FAQ
Do I need a paid Claude plan?
Yes. Claude Projects are a Claude.ai Pro, Team, or Enterprise feature.
What should I attach to the knowledge base?
Start with llms-full.txt (the complete Convex corpus). Optionally add the three composite methodology papers as separate files for high-priority retrieval. File size is within Claude Project limits.
Can Claude call the Convex API from inside a Project?
Projects do not call external APIs directly, but if you enable the Convex MCP server in Claude Desktop, the Project inherits those tools for tool-use sessions. Install instructions are at convextrade.com/mcp.
How do I keep the knowledge base fresh?
The llms-full.txt endpoint regenerates daily. You can re-upload periodically or script a refresh. For strictly live data, use the MCP server instead of the knowledge attachment.