MCP Server
The REST API is also available over the Model Context Protocol , so AI agents and MCP-compatible clients can call the wstGBP reads as tools instead of raw HTTP. The tools cover nearly the full token surface: live state with reserves, quotes, bridge status, events, and history. Only the token list and the CoinGecko-compatible plain-text supply routes stay REST-only. The server is public and read-only, speaking stateless JSON-RPC 2.0 over the Streamable HTTP transport (there is no SSE endpoint):
https://api.arb.capital/api/mcp/mcpTools
| Tool | Arguments | Returns |
|---|---|---|
get_token_stats | token | Live wrapper state, trailing APR, and a reserves/solvency block, derived from /api/tokens/{token}, /reserves, and /apr |
get_token_daily | token, days (default 30, 1-365) | Daily NAV / supply / TVL history, derived from /daily |
get_token_snapshots | token, days (default 7, 1-30) | Raw hourly NAV / supply / TVL snapshots, derived from /snapshots |
get_price_history | token, days (default 90, 1-365) | Oracle poke history including pause events, derived from /price-history |
get_token_events | token, type, user, days (default 30, 1-365), limit (default 50, 1-200) | Mint / redeem / claim events, newest first, derived from /events |
get_quote | token, side (mint | redeem), amount (wei) | Read-only mint/redeem preview with executability verdict, derived from /quote |
get_bridge_status | token | Bridge deployments merged with per-chain supply and staleness, combining /chains and /supply |
get_pools | token | Live Uniswap v4 venue state, derived from /pools |
get_pool_history | token, pool (key or poolId), days (default 7, 1-30) | Hourly snapshots for one pool, derived from /pools/history |
get_ops_status | none | Freshness of the snapshot pipeline behind the history routes |
get_docs | page (optional) | This documentation site, fetched live: the page index with no arguments, or one page as markdown (for example guides/mint-redeem) |
The tools read from the same data layer as the REST routes and follow the same
conventions: token accepts the slug or home-chain
address, wei amounts are strings, prices are WAD, and Float fields carry
display-precision copies. Result shapes are close to the linked REST bodies but not
identical: tools echo token (and their window arguments) alongside the rows;
get_token_stats groups APR under an apr key and adds the reserves block; and
some history tools use shorter default windows or stricter row limits than their REST
counterparts; prefer REST for bulk pulls. No API key is needed. The endpoint shares
the 30 req/min live-read rate limit and, unlike the REST routes, is never
edge-cached, so every tool call counts against that limit. Bad input or a failed
read comes back as a tool result flagged as an error, with a plain-text message
(unknown token, out-of-range amount, unknown pool key); problems below the tool
layer, such as an exceeded rate limit or a malformed request, surface as plain HTTP
errors instead, most commonly 429 with a Retry-After header, as on the REST
routes.
The server is built for integrators: every successful tool result except
get_ops_status and get_docs (which have no REST equivalent) carries a
restRoutes field listing the REST call(s) equivalent to the exact arguments used,
so an agent that explored the data over MCP can transcribe the same reads straight
into application code. The quote, pool, and bridge tools also return a docs field
linking the relevant guide on this site. The server’s initialize instructions
point the same way, and two documents are exposed for clients that support MCP
resources: the full REST reference as api-reference, and the complete content of
this site as integration-docs.
Machine-readable docs
This site publishes plain-markdown mirrors of every page, regenerated on each deploy, following the llms.txt convention :
https://docs.wstgbp.com/llms.txt page index
https://docs.wstgbp.com/llms-full.txt every page in one markdown file
https://docs.wstgbp.com/index.html.md homepage
https://docs.wstgbp.com/<page>.md one page, e.g. /guides/mint-redeem.mdThe get_docs tool and the integration-docs resource read these files at request
time, so what an agent sees always matches the deployed site.
Connect a client
To add it to Claude Code:
claude mcp add --transport http arb-api https://api.arb.capital/api/mcp/mcpAny client that supports remote MCP servers over Streamable HTTP connects the same way, including claude.ai custom connectors and the Claude API MCP connector.