agent-graph-mcp
MCP server for graph-orchestrated LLM workflows — 25 typed tools, daemon/proxy architecture, checkpoint/resume, human-in-the-loop approvals, and HMAC-authenticated execution receipts.
What it gives you
- 25 typed MCP tools — graph lifecycle, execution (sync + async), state inspection, checkpoint/resume, HITL approvals, source witnesses, templates, policy validation
- Daemon + proxy architecture — single-process daemon with file lock ownership, crash recovery, and startup mode enforcement; stateless proxy that bridges stdin/stdout to Unix socket
- Durable persistence — SQLite-backed with atomic checkpoint transactions, no partial rows after crash
- Deterministic local checkpoint/resume — HMAC-SHA256 authenticated checkpoints for linear chains of deterministic
passthroughandstate_transformnodes - Built-in templates —
council_deliberation(3-analyst parallel),parallel_council(debate),plan_critique_refine,analysis_pipeline,classifier_router - Evidence witnessing — caller-supplied source capture with HMAC-authenticated receipts
Quick start
npx (recommended)
Cargo install
Hermes config
mcp_servers:
agent_graph:
command: npx
args:
- -y
- "@recursiveintell/agent-graph-mcp"
- --direct
- --base-url
- http://127.0.0.1:11434
- --model
- glm-5.2:cloud
Tools (25)
Graph lifecycle: graph_create, graph_list, graph_inspect, graph_render
Execution: graph_execute, graph_run_start, graph_run_wait, graph_run_cancel, graph_run_get
State: graph_run_state, graph_run_events, graph_run_checkpoint, graph_run_resume
Approval: graph_approval_list, graph_approval_get, graph_approval_request
Evidence: graph_source_witness_capture, graph_source_witness_get
Templates: graph_template_list, graph_template_instantiate, graph_template_candidates, graph_template_outcomes
Policy: graph_policy_check
Receipt: graph_run_receipt
Status: graph_status
Built-in templates
| Template | Description |
|---|---|
council_deliberation |
3-analyst parallel council: coordinator → fanout → researchers → join → synthesize |
parallel_council |
2-person debate: optimist vs skeptic → join → judge |
plan_critique_refine |
Sequential plan → critique → refine |
analysis_pipeline |
planner → researcher → extractor → synthesizer → validator with correction loop |
classifier_router |
LLM classifier routes input to bug/feature/question handlers |
Verification
# npx smoke test
| \
| \
# Expected: 25
# Build + test
Ecosystem
| Crate | Description |
|---|---|
| agent-graph-mcp | MCP server (this crate) |
| ri-agent-graph | Core graph execution engine |
| llm-pipeline | Reusable LLM node payloads |
| stack-ids | Trace/identity primitives |
License
MIT © RecursiveIntell