Serveur MCP ReadyAgents

ReadyAgents Core

ReadyAgents runs YAML agent workflows locally. Every node is checkpointed, so a run pauses for human approval, resumes from where it stopped, and replays offline as a regression test. Your keys, your machine, no daemon.

Site: readyagents.dev. Repo: github.com/readyagentsdev/readyagents-core.

Tried it? Open an I-ran-this issue. We are not launching. We are listening.

This repository is the free core. You keep the provider account and the bill. Install with pip install readyagentsdev, or from this clone.

60-second start

Requires Python 3.11–3.14 on Linux, macOS, or Windows. Current version is 2.0.12. Install with pip install readyagentsdev, or from this clone. The 2 versions the core contract only; extras carry their own maturity tiers.

pip install readyagentsdev
readyagents new my-flow
readyagents run my-flow/workflow.yaml
readyagents runs list

The wheel ships the example workflows too — no clone needed:

readyagents new f --from-example calc_pipeline
readyagents run f/workflow.yaml

readyagents new --list-examples shows all of them.

Or from a clone:

git clone https://github.com/readyagentsdev/readyagents-core.git
cd readyagents-core
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e .
readyagents run examples/calc_pipeline.yaml
readyagents runs list
readyagents doctor

readyagents run examples/calc_pipeline.json is the same graph.

HITL next: docs/first-ten-minutes.md.

What it does

• Define agent workflows as YAML or JSON (nodes + edges)
• Run agent, tool, condition, transform, approval, parallel, include, foreach, and decide nodes. Agent nodes may declare a tools: allowlist for a bounded tool-use loop. type: decide is experimental; see docs/decisions.md.
• Persist after every node and resume a paused or failed run from the last successful node
• Inspect past runs: readyagents runs list / show / replay / fork / diff / freeze / report (local HTML)
• Scaffold a starter: readyagents new my-flow (basic, approval, research, pipeline, review, foreach, agent-tools, gated, decide)
• Builtin tools with zero extra servers: now, calc, json_get, list_dir, read_file, write_file
• MCP client and server (readyagents mcp serve, readyagents mcp probe) with official tasks and MRTR approvals
• Policy, spend ledger, budgets and caps: per-node token/cost, --estimate / --max-spend caps, model fallback, JSON logs

Beyond the core, 40+ opt-in extras ship in the same install — teams, studio, browser, knowledge, distillation, registry, environments and more. Each is listed with its maturity tier and its limits in docs/extras.md. None of them is required, and none of them changes how the core behaves.

The agent firewall (taint, tool policy, MCP pinning — security model, policy) is defence in depth, not a solution to prompt injection.

Architecture

flowchart LR
  YAML[Workflow YAML/JSON] --> Engine
  subgraph Core["ReadyAgents Core"]
    Engine[Workflow engine]
    Tools[Builtin tools]
    LLM[BYOK LLM providers]
    MCP[MCP client / server]
    Packs[Pack loader]
  end
  Engine --> Tools
  Engine --> LLM
  Engine --> MCP
  Packs --> Engine
  Packs --> Tools
  LLM --> OpenAI[OpenAI]
  LLM --> Anthropic[Anthropic]
  LLM --> Compat[OpenAI-compatible]

CLI

Command	Purpose
readyagents init	Write .env from .env.example if missing
readyagents new [name] [--template basic|approval|research|pipeline|review|foreach|agent-tools|gated|decide] [--from-example NAME] [--list-examples]	Scaffold workflow + README + .env.example + local JSON Schema, or copy a shipped example
readyagents import SOURCE PATH [--out DIR] [--explain SOURCE]	Import n8n / LangGraph / CrewAI / trigger-action exports (structural translation only)
readyagents validate PATH	Schema-validate a workflow (source-located errors on failure)
readyagents schema	Print/write/check the generated workflow JSON Schema
readyagents eval PATH	Score a keyless fixture suite (exit 0/1)
readyagents simulate PATH [...]	Declaration-driven cases scored with eval; distinct failures frozen
readyagents optimize PATH --eval SUITE	Reflective prompt optimization against your eval suite
readyagents prompts list / show / history / diff / rollback	Versioned prompts beside the workflow; rollback restores exactly
readyagents env status / history / diff / deploy	Declared environments and pinned releases
readyagents registry scan / annotate / check / list / show / stats / card / export	Local agent inventory from declared roots (Annex VIII export is a draft)
readyagents distill plan / dataset / train / evaluate / promote	Local one-node adapters from consented runs (pack trains; holdout required)
readyagents promote PATH --from SRC --to DST	Copy a source pin onto a target after eval/fixture/bench/health/approval gates
readyagents rollback --env NAME	Restore the previous release atomically; never auto-forwards
readyagents feedback export / stats	Consent-gated correction datasets (production data)
readyagents run PATH [--input KEY=VALUE] [--dry-run] [--approve NODE] [--reject NODE] [--decision-file FILE] [--actor NAME] [--pack PATH] [--policy PATH] [--estimate] [--max-spend USD] [--max-tokens N] [--label KEY=VALUE] [--sovereign] [--stream] [--env NAME]	Execute (or --estimate without running). --env runs a pinned release. --stream is opt-in
readyagents batch PATH --input-file FILE [--concurrency N] [--max-spend USD] [--out FILE]	Foreground: one workflow, many JSONL/CSV rows (opt-in)
readyagents attest RUN_ID	Data-residency attestation (technical evidence, not legal compliance)
readyagents bundle --out DIR	Offline wheel set for pip install --no-index --find-links
readyagents resume RUN_ID [--approve NODE] [--reject NODE] [--decision-file FILE] [--policy PATH]	Resume a paused or failed run
readyagents wake	Evaluate wait conditions (lazy; starts no timer or daemon)
readyagents event	Inject a signed event (unsigned events are refused)
readyagents policy check PATH	Validate a firewall policy file (fail closed)
readyagents policy explain PATH [--policy PATH]	Show which tools each node may call and why
readyagents evidence RUN_ID [--out DIR]	Local evidence pack
readyagents audit verify [--file PATH]	Walk the hash-chained audit trail
readyagents spend [--since DATE] [--by day|workflow|model|actor|label]	TokenOps: aggregate the local spend ledger (informational vs the provider invoice)
readyagents graph PATH	Deterministic Mermaid routing (executes nothing)
readyagents decide RUN_ID [--file FILE | --node ID --decision approve] [--token-file JWT] [--actor NAME] [--reason TEXT]	Inject an approval; --actor stays the default, --token-file identifies
readyagents approvals list [--role ROLE] [--actor NAME] [--expiring-within 1h]	Queue of paused gates the caller may see
readyagents delegate --from A --to B --until TS [--scope ROLE]	Time-bounded, single-hop, revocable approval delegation
readyagents delegations list | revoke ID	List or revoke local delegations
readyagents identity verify --token-file JWT	Verify an assertion against local trust anchors
readyagents identity whoami	Workload fingerprint (never the private key)
readyagents runs list	List persisted runs
readyagents runs show RUN_ID	Node timeline + stored state (inspect is an alias)
readyagents runs report RUN_ID	Local HTML summary of a run
readyagents runs replay RUN_ID	New run from stored inputs
readyagents runs delete RUN_ID --yes	Delete one local run record
readyagents runs gc --yes	Prune succeeded/failed/cancelled runs (paused kept; in-window records refused unless --override-retention)
readyagents runs fork / diff / freeze / migrate	Time machine: fork a run, diff two, freeze a cassette, migrate JSON→SQLite
readyagents connectors list / show / test	Small governed catalog (rest, sql, object_storage, message, ingest) — not 500 SaaS
readyagents sign / verify / lock / sbom / trust	Supply-chain: signatures prove origin, not safety
readyagents approvals serve	Foreground localhost approval page
readyagents studio [--port 8790] [--open] [--read-only]	Foreground localhost canvas and run inspector
readyagents models list / show / route --explain	Dry model catalog and routing explain (no provider call)
readyagents mcp serve	Stdio MCP server (builtin tools); --json prints protocol versions
readyagents mcp probe URL	Read-only server/discover diagnostic (never calls a tool)
readyagents a2a serve PATH	Foreground A2A door for one workflow (loopback by default)
readyagents a2a card PATH	Deterministic Agent Card (no network)
readyagents a2a probe URL	Read-only remote card diagnostic (no secret values)
readyagents memory list / show / search / forget / export	Local scoped memory (offline except optional embeddings)
readyagents knowledge ...	Ingest, sync, cite, and forget knowledge documents (foreground only)
readyagents table ...	Inspect intermediate tables: head, schema, stats
readyagents triggers ...	Inspect declared triggers, dry-run mappings, list events (starts no listener)
readyagents skills ...	Install, list, export, and remove Agent Skills (open format; no marketplace)
readyagents agents-md	Emit project context: how to run, validate, and test workflows here
readyagents package ...	Build, install, and catalog workflow packages (review-before-install)
readyagents health	Cluster failures by fingerprint over the run store (no daemon, no telemetry)
readyagents bench ...	Offline-by-default benchmark suite with labelled engine vs live timing
readyagents sessions ...	List, show, close, replay, and freeze conversational sessions (turns are runs)
readyagents serve ...	Foreground loopback surfaces (stops with the process)
readyagents packs [--pack PATH]	List installed / local packs
readyagents doctor	Read-only platform / extras / permissions / loopback / run-store / sovereign diagnostic
readyagents version	Print version

Examples (no keys unless noted)

File	What it shows
examples/calc_pipeline.yaml	Builtin tools, transform, condition
examples/calc_pipeline.json	Same graph as calc_pipeline.yaml
examples/approval_gate.yaml	Human-in-the-loop pause / resume
examples/ollama_local.yaml	Keyless loopback OpenAI-compat path (no live model)
examples/quorum_gate.yaml	Two-approver gate (keyless)
examples/expiring_gate.yaml	Lazy deadline, on_expire: reject (keyless)
examples/multi_gate.yaml	Two sequential approval gates
examples/fanout_gate.yaml	Parallel branches + approval
examples/include_demo.yaml	Sub-workflow include (new --from-example)
examples/composed_gate.yaml	Include + parallel + approval (new --from-example)
examples/research_brief.yaml	Agent node (needs a key)
examples/support_triage.yaml	Classify then branch (needs a key)
examples/code_review.yaml	read_file + review (needs a key)
examples/agent_tools.yaml	Agent tools: [calc] (needs a key; --dry-run is keyless)
examples/foreach_calc.yaml	Sequential foreach + calc (no keys)
examples/policy_gated.yaml	Policy gate on tainted write_file (no keys)
examples/readyagents.policy.yaml	Starter firewall policy
examples/json_mutate.yaml	json_set / json_merge (no keys)
examples/table_pipeline.yaml	type: table + type: classify (validate keyless)
examples/list_dir.yaml	Builtin list_dir (no keys, no MCP, no Node)
examples/eval/pass.yaml	Keyless readyagents eval fixture suite
examples/a2a_delegate.yaml	type: a2a dry-run (no network)
examples/memory_triage.yaml	type: memory write then search (keyless)
examples/code_reshape.yaml	type: code JSON in/out (keyless, subprocess)
examples/batch_echo.yaml	Keyless readyagents batch row echo (batch_rows.jsonl / .csv)
examples/connector_rest.yaml	Keyless rest connector against a local fixture
examples/connector_demo.yaml	Local --pack connector (examples/packs/connector_pack.py)
examples/gated_write.yaml	Approval then write_file (no keys)
examples/env/echo.yaml	Keyless env deploy + run --env (examples/env/readyagents.env.yaml)

Docs

• Why ReadyAgents?
• Getting started
• First ten minutes
• Concepts
• Configuration (BYOK)
• Workflows
• Authoring (JSON Schema, editors, located errors)
• Connectors (small catalog; write-shaped ops gate by default)
• Sovereign mode (in-process egress refuse, not an OS sandbox)
• Cost / TokenOps (informational vs the provider invoice)
• Time machine (record / replay / fork / diff / freeze)
• MCP · MCP paste catalog
• A2A (untrusted remote content; delegation can exfiltrate; not certification)
• Memory (untrusted; delayed injection and scope escape)
• Scale and batch (opt-in foreground batch; benchmarks labelled engine vs live)
• Streaming (opt-in --stream; not audio)
• Guardrails / output contracts (opt-in contract:; declared rules)
• Multi-agent teams (opt-in type: team; closed members; no quality claim)
• Model routing (opt-in routing:; declared policy)
• Multimodal I/O (opt-in MediaPart / type: document / type: transcribe; extras; not an OCR claim)
• Knowledge pipelines (opt-in type: ingest; citations/freshness)
• Data pipelines (opt-in type: table / type: classify)
• Long-horizon waits (opt-in type: wait; lazy wake)
• Event triggers (opt-in triggers:; no listener in core; at-least-once plus idempotency)
• Agent Skills (opt-in type: skill; open format)
• Governed browser use (opt-in type: browser; declared actions; no CAPTCHA; optional pack)
• Conversational sessions (opt-in type: converse; loopback serve chat; no audio in core)
• Environments and rollout (opt-in readyagents.env.yaml; pinned releases; gated promote)
• Migration (opt-in readyagents import; structural translation only)
• Agent registry (opt-in readyagents registry; declared roots; draft Annex VIII export)
• Distillation (opt-in readyagents distill; pack-owned training; holdout-gated adapters)
• Packaging (opt-in readyagents package; review-before-install)
• Simulation (opt-in readyagents simulate; declaration-driven cases)
• Self-healing (opt-in readyagents health / recovery:; fingerprints and fail-safe gates)
• Benchmarks (opt-in readyagents bench; offline cassettes)
• Packs
• Supply-chain trust (signatures prove origin, not safety)
• Continuous pack (optional, separate distribution)
• Platform support
• CLI
• Compliance evidence (Articles 12–14 mapping; not certification)
• Observability
• Changelog
• Release notes 2.0.12

Install extras

LLM and MCP extras are optional.

pip install "readyagentsdev[openai]"
pip install "readyagentsdev[anthropic]"
pip install "readyagentsdev[gemini]"
pip install "readyagentsdev[bedrock]"
pip install "readyagentsdev[vertex]"
pip install "readyagentsdev[mcp]"
pip install "readyagentsdev[all]"
pip install "readyagentsdev[image]"
pip install "readyagentsdev[pdf]"
pip install "readyagentsdev[audio]"
pip install "readyagentsdev[table]"

From a clone, the same extras are pip install -e ".[openai]" (and anthropic / gemini / bedrock / vertex / mcp / all). The optional [otel] extra is not included in [all]; it starts no collector (see observability.md). The optional [sign] extra (Ed25519 artifact signatures), [jwt], [gemini], [bedrock], [vertex], [image], [pdf], [audio], and [table] extras are also not in [all]. Unsigned default runs never import them. Codecs are extras; core installs stay text-only. Pandas is optional for table ops; stdlib is sufficient.

Then cp .env.example .env and paste your own keys. Core workflows that only use builtin tools do not need extras, keys, or Node.js.

docker compose run --rm readyagents run examples/calc_pipeline.yaml
make smoke

What is not in this repository

Always-on packs are waitlisted and not for sale.

Always-on / continuous workers are not in Core. The optional readyagents-pack-continuous distribution (separate repository, not a Core extra) can run configured workflows from an explicit foreground command. Installing Core still starts no scheduler or listener.

Hosted control plane. Hosted recovery and remote run stores. SSO, multi-tenant teams, billing.

The core has persist, resume, and approval pauses for a local one-shot. It does not run always-on.

License

Apache License 2.0. See LICENSE.

Security

Please report vulnerabilities as described in SECURITY.md. Public contact: info@readyagents.dev. Do not commit API keys. Local operator files such as .env are gitignored.