ChainWeaver
Expose pre-defined ChainWeaver flow files and Python tools as deterministic MCP tools so agents run multi-step pipelines without an LLM between every step.
Overview
io.github.dgenio/chainweaver is a Build-phase MCP server that exposes deterministic ChainWeaver flows and Python tools to agents without LLM calls between steps.
What is this MCP server?
- Serves a .flow.yaml or .flow.json file as MCP tools via uvx chainweaver serve
- Deterministic steps with no LLM calls between ChainWeaver flow stages
- Repeatable --tools flags to mount Python modules that export Tool instances
- PyPI package chainweaver v0.12.1 with chainweaver[mcp] extra for fastmcp dependencies
- PyPI package chainweaver version 0.12.1
- stdio transport via serve subcommand
- Requires filepath to .flow.yaml or .flow.json
Community signal: 6 GitHub stars.
What problem does it solve?
Your agent workflows waste tokens and drift when every pipeline step goes back through the model instead of running fixed logic.
Who is it for?
Builders who already define ChainWeaver flows and want Claude Code or Cursor to trigger those pipelines as first-class MCP tools.
Skip if: Anyone who only needs one-off REST calls or fully improvisational agent reasoning with no authored flow graph.
What do I get? / Deliverables
After registration, your agent can call stable MCP tools backed by flow files and Python Tool modules with predictable intermediate behavior.
- MCP tools mapped from a ChainWeaver flow definition
- Agent-triggerable deterministic pipelines without per-step LLM round trips
Recommended MCP Servers
Journey fit
ChainWeaver sits in Build because you configure flows and tool modules while shaping how your agent executes reliable automation. Agent-tooling is the right shelf for MCP servers that wrap orchestration logic and make coding agents call structured, repeatable workflows.
How it compares
Deterministic flow orchestration MCP server, not a marketplace skill or a hosted no-code automation UI.
Common Questions / FAQ
Who is io.github.dgenio/chainweaver for?
Developers shipping agent workflows who want ChainWeaver .flow files and Python tools exposed deterministically over MCP.
When should I use io.github.dgenio/chainweaver?
Use it when you have repeatable multi-step automations that should run as tools without an LLM between each ChainWeaver step.
How do I add io.github.dgenio/chainweaver to my agent?
Run uvx with --from chainweaver[mcp], pass serve and the path to your flow file, add optional --tools module paths, and register that stdio command in your MCP client config.