Undertow — market liquidity map

What Undertow — market liquidity map does

Undertow provides read-only market-liquidity research through MCP. Its public surface focuses on BTC exit analysis, venue depth, liquidity conditions, and supporting evidence. The service can compare estimated sell impact at published size rungs, show concentration across venues, identify historical depth-collapse episodes, and describe the coverage and rights behind market observations.

The broader hosted catalog contains 24 tools: 16 public tools and eight subscriber tools. Anonymous access exposes the public inventory, while subscriber eligibility is checked for restricted requests. The local stdio adapter exposes the 16 anonymous tools and three prompts.

Use Undertow for research before an agent evaluates a hypothetical exit, compares venues, or combines market-liquidity evidence with funding and institutional context from related products. It is not an order router, execution system, live trading interface, or source of trade clearance.

How it works

The primary endpoint is a hosted Streamable HTTP MCP service at https://api.seiche.info/undertow/mcp. Public read-only calls do not require an API key, wallet, or account. A request can return published venue estimates, snapshot times, source clocks, conversion details, excluded costs, and reasons for withheld results.

BTC exit tools use stated sell sizes and published depth information. crypto_exit_check reports four BTC sizes by rung or bracket, while exit_cost estimates impact using interpolated bid-depth bands. Fees are excluded. The safety context tool can provide exact-rung BTC/USD paper-observation details, including a request hash, point-in-time source and rights information, venue clocks, conversion, bid bands, and startup revision.

Other tools calculate input-based funding carry, linear portfolio shocks, gold cash-realisation arithmetic, venue concentration, and BTC/ETH price reconciliation. These calculations preserve assumptions rather than presenting forecasts or execution instructions. Research calls do not place trades.

Setup and configuration

The hosted connection requires no local installation:

claude mcp add --transport http undertow https://api.seiche.info/undertow/mcp

Claude Code, Cursor, and VS Code setup references are available through the product site, and clients that support custom MCP connectors can use the same URL. For clients that require stdio, clone the repository at a reviewed commit, install Python 3.12 or newer and uv, then run:

uv sync --locked
uv run --locked undertow-mcp

A Docker image can also run the stdio adapter. It needs outbound HTTPS access to api.seiche.info; no API key, bearer token, wallet, port, volume, or environment setting is required. The adapter blocks arbitrary upstream URLs and redirects, and preserves upstream errors and rights refusals.

Tools and capabilities

• Check access tier, entitlement, usage, and subscriber routes with agent_access_status.
• Estimate BTC sell impact, compare venues, and inspect excluded or withheld data with crypto_exit_check and exit_cost.
• Review funding carry, linear portfolio shocks, and concentration across assets or venues.
• Inspect depth-collapse episodes, liquidity tiers, corpus coverage, research metadata, and sealed outcome records.
• Retrieve the latest published article and its source clock or quality receipt.
• Examine BTC/ETH venue-price reconciliation and paper-observation context without enabling live execution.

Limitations and notes

Estimates are evidence for research, not executable quotes. Fees are excluded from BTC exit-cost results, and venue rankings may change between snapshots. Coverage can be withheld when data is stale, future-dated, uncovered, unconverted, unavailable, or fails replay or contract checks. A zero depth-episode value is distinct from missing data.

Funding calculations use a constant-rate scenario rather than a forecast. Portfolio stress applies linear input shocks and does not model liquidation or probabilities. Gold cash-realisation performs arithmetic from supplied inputs; it does not fetch prices, verify balances, store assets, or execute sales. Venue reconciliation is experimental and has no execution authority. Keep observation dates, venue and size assumptions, rights, and coverage gaps with every result.