Skip to content

feat: add AlgoVault MCP integration quickstart - #6

Open
AlgoVaultFi wants to merge 2 commits into
crewAIInc:mainfrom
AlgoVaultLabs:algovault-mcp-quickstart
Open

feat: add AlgoVault MCP integration quickstart#6
AlgoVaultFi wants to merge 2 commits into
crewAIInc:mainfrom
AlgoVaultLabs:algovault-mcp-quickstart

Conversation

@AlgoVaultFi

Copy link
Copy Markdown

What this adds

A new MCP Integration/ quickstart notebook, algovault_mcp.ipynb, showing how to connect CrewAI to a remote MCP server with MCPServerAdapter over the streamable-http transport.

The example server is AlgoVault MCP: it returns a composite BUY/SELL/HOLD trade call across 5 crypto perp venues, with a track record that is Merkle-anchored on Base L2 (ERC-8004 agentId 44544).

Why it is not just another "MCP + framework" demo

The notebook is built around what is specific to the data, not the plumbing:

  • Composite verdict, not raw signals. One get_trade_call response gives call / confidence / indicators / regime — no client-side vote-counting over separate indicators.
  • Cross-venue. Signal is fused across Binance, Bybit, OKX, Bitget, and Hyperliquid — not reproducible from any single-venue endpoint.
  • On-chain verifiable. Each signal is anchored on Base L2, so the track record can be checked at algovault.com/track-record rather than taken on trust.

Notebook contents

  1. Intro + prerequisites (Python 3.10+, pinned crewai / crewai-tools[mcp]).
  2. Connect with MCPServerAdapter and list the AlgoVault trade tools.
  3. Read a live trade call with no LLM key — the executed output cell shows a real verdict.
  4. A note on the adapter's null handling: pass concrete values for optional arguments (includeReasoning, exchange, assetClass) so they are not forwarded as null.
  5. An optional, key-gated Crew + Agent wiring that is skipped when no provider key is set.

The output cells were executed end-to-end against the live server. The notebook is pre-sanitized with scripts/clean_notebooks.py, so the Clean Notebooks check is a no-op.

Conventions followed

  • New top-level feature directory (MCP Integration/), matching the existing per-feature layout (Collaboration, Custom LLM, Guardrails, …).
  • nbformat 4, python3 kernelspec, %pip install install cell — matching the existing notebooks.
  • Verified against crewai==1.15.1 + crewai-tools[mcp]==1.15.1 on Python 3.12.

Further reading

Happy to adjust the directory name or notebook structure to match maintainer preference.

Megatron888-Robot and others added 2 commits July 1, 2026 12:22
Adds a new "MCP Integration" quickstart notebook demonstrating how to
connect CrewAI to a remote MCP server via MCPServerAdapter over the
streamable-http transport. The example uses AlgoVault MCP, which returns
a composite BUY/SELL/HOLD trade call across crypto perp venues.

The notebook loads the tools, reads a live trade call with no LLM key,
documents the concrete-default gotcha for the adapter's null handling,
and includes an optional key-gated Crew wiring. Output cells are executed
against the live server.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Track-record snapshot in the intro cell refreshed from the live
/api/performance-public + /api/merkle-batches endpoints (calls, PFE win
rate, on-chain batch count, asset coverage). The notebook already tells
readers to verify live at algovault.com/track-record; this keeps the
static snapshot line current.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants