MCP server for orchestrating parallel Claude agent workloads via Docker containers
Wellknown found it in public sources; nobody has proven control of it yet. Claiming takes one click if the repository is under your GitHub account, or a small file on your domain otherwise. Verified owners get the badge, 15-minute checks, status alerts, edits that outrank crawled data, and a ranking boost.
Agents can do it too: POST https://wellknown.network/api/v1/claims with {"agent":"mcp-swarm","method":"well_known_file"} — machine-readable steps at claim.json, guide at /docs/claim.
Everything here was measured by our prober or read from a registry. Nothing is self-reported.
Attributed to the source that supplied each field. Treated as claims, not facts.
<!-- mcp-name: io.github.stiege/mcp-swarm --> # swarm-mcp **Orchestrate parallel Claude agent workloads via Docker containers.** [](https://pypi.org/project/mcp-swarm/) [](https://python.org) [](LICENSE) [](https://docker.com) [](https://stiege.github.io/swarm-mcp/) swarm-mcp is an MCP server that lets your Claude session spawn other Claude agents — each in an isolated Docker container — and compose their results using functional combinators. Instead of one agent doing everything, you describe the work as `run`, `par`, `map`, `chain`, `reduce`, and `pipeline` calls, and swarm-mcp handles container lifecycle, resource scheduling, and result plumbing. The agent outputs are lazy refs (metadata on the wire, text on disk) so you never blow up the MCP protocol with megabytes of agent output. --- ## Why swarm-mcp? - **True Docker isolation per agent.** Every agent gets its own container with its own filesystem, network policy, memory limit, and CPU quota. No shared state leaks between agents. A rogue `rm -rf /` in one container doesn't touch anything else. - **Monadic ref architecture (lazy by default).** Combinators return refs — small JSON objects with metadata (cost, duration, exit code, provenance hash). The actual text stays on disk in `/tmp/swarm-mcp/{run_id}/{agent_id}/result.json`. Call `unwrap` when you actually need the content. This keeps the MCP protocol fast and prevents context window bloat. - **Free monad pipelines (computation as data).** The `pipeline` tool interprets a JSON definition — steps with `on_fail` handlers, `condition` guards, `retry_if` loops, and `next` jumps. The pipeline is data you can store, versi…
Mapped onto the structured taxonomy from declared text and observed tool names. Confidence shown for derived entries.
Every source is kept verbatim. Field changes are logged as events.