What it does
Jiheng is an exact math execution layer for AI agents and MCP clients. It covers fractions, roots, symbolic expressions, statistics, finance, units and dates. Where possible it returns exact values, such as 1/3 kept as a fraction or sqrt(2) kept as a root, and it states clearly when an exact result is not possible. It does not round silently or guess units, and its error messages are written so that a model can correct its own call. It can run as an MCP server or be attached as a DSH plugin.
Developers building AI agents or MCP clients can use it when a model needs precise amounts, units, dates or statistical calculations, for example in finance, quantitative, engineering or data agents. It does not read problems, build models or reason about them, and it does not retrieve data. Live market data, large CSV files or database results should be gathered first by other tools and then passed to it for calculation.
Running it requires uv. The MCP client is configured to launch the server with uvx, which downloads jiheng-mcp on first start and uses a local cache afterwards. No API key is needed. A resident worker process gives a publisher-reported latency of about 3 ms per call, and an optional worker pool can handle parallel batches. Policy settings control when the model calls the tool, and limits control series length, expression size and timeouts.
Limitations and requirements
Requires uv to be installed; the first launch downloads jiheng-mcp. Does not read problems, build models or reason. Not a replacement for Wolfram-level symbolic integration or PDE solving; symbolic depth is described as sufficient but not Wolfram-level. Does not fetch live market data, read large CSV files, connect to databases, scrape web pages or read files. calc_shard suits only decomposable problem types such as interval integration, interval sums and parameter sweeps. Default single-evaluation timeout is 8000 ms.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.CQQOOQ123/jiheng
Exact math execution layer for AI agents: fractions, finance, units, dates. 19 packs.
v0.3.9As shown on registry.modelcontextprotocol.io when we last read it