Monarchic LLC
Autonomous engineering
MCP layer available first
Autonomous engineering as a service
Monarchic is building the platform for agent teams that plan, code, verify, remember, and ship work with evidence. The first commercial surface is the MCP layer: ExplicitMem for durable memory, RepoIntel for repository intelligence, BrowserOps for browser evidence, and proof tools for release-grade handoff.
Platform Direction
Agents that can be trusted with real engineering work.
Now
Hosted MCP tools for memory, repo context, browser checks, release evidence, and agent handoffs.
Next
Account-owned autonomous engineering workflows with bounded tools, durable state, and reviewable proof.
Rule
No invisible agent work. Benchmarks, screenshots, logs, and receipt files stay attached to the decisions they support.
Commercial surface
The MCP catalog is the first layer of the larger Monarchic platform.
View catalogWho It Is For
Teams preparing agents to do real engineering work without losing memory, proof, or control.
Account scoped
Routes and API keys belong to an account, not a loose demo token.
Proof included
Research and receipts say what each MCP proves and what it does not.
Coming soon
We are holding access until the account controls are boring and reliable.
01 / ExplicitMem
Durable memory with receipts
99.8% benchmark accuracy, graph memory, bounded recall, and audit-friendly retrieval.
02 / RepoIntel
Repository intelligence for agents
59 evaluated rows with citations, dependency summaries, and targeted code evidence.
03 / BrowserOps
Browser verification evidence
Screenshots, console diagnostics, network checks, and scenario receipts.
Monarchic Vision
Autonomous engineering as a service, built from proof-carrying tools.
The destination is not a pile of MCP servers. It is an account-owned engineering platform where agents remember context, inspect code, operate browsers, produce receipts, and hand off work that humans can verify.
Persistent state
ExplicitMem keeps useful memory outside hidden chat history and makes recall inspectable.
Repository ground truth
RepoIntel gives agents source paths, citations, symbols, and risk notes instead of vague summaries.
Browser evidence
BrowserOps captures what happened in the UI with hashes, screenshots, console output, and verification receipts.
Release proof
ProofPack, ReleaseOps, and Verified turn tests, patches, packs, and release gates into reviewable artifacts.
Repository Demo
Ask for repo facts. Get structured context, source paths, and evidence an agent can reuse.
This is the kind of output RepoIntel is built for: not a vague summary, but a bounded readout of architecture, dependencies, entry points, and files that prove the answer.
repointel / live repository extract
streaming repo facts into agent context
Index
Evidence
Handoff
$ monarchic repointel inspect ./monarchic-webapp \
--question "where does checkout data come from?"
_
indexed 579 components / 6854 files
found Stripe checkout surface
linked source paths + evidence notes
warn unresolved ownership boundaryExtracted Facts
- Checkout behavior is traced to route handlers, catalog data, and UI entry points.
- Every claim includes source paths an agent can open during the next task.
- Risk notes flag unknown owners, missing evidence, and stale assumptions.
Evidence paths
14
Confidence
High
Agent handoff
Export the repository findings as structured context with source paths, confidence, and evidence notes an agent can reuse in the next task.
InfraProfiler MCP
Ask what infra is hurting reliability, cost, or deployment speed.
InfraProfiler already caught a real Monarchic staging failure: an idle worker burning CPU because SQS was being polled hot. The agent traced it to owner repos, fixed the queue behavior, and added a release guard without scaling up compute.
CPU max
98.45%
Queue wait
0s -> 20s
Task size
+0
Guard
live
infraprofiler / hotspot report
aws staging profile / no mutation
$ infraprofiler rank_hotspots \
--provider aws \
--env staging \
--max-resources 20
_
detect worker CPU max 98.45%
compare queue depth stayed empty
rank SQS hot polling as root target
guard long polling required at deployECS service
monarchic-staging-worker
hot polling
Owner
monarchic-worker
SQS queue
monarchic-staging-queue
empty receives
Owner
monarchic-infra
release guard
worker-sqs-long-polling-check
regression block
Owner
monarchic-plan
Detect
Staging worker hit 98.45% max CPU while the queue had no meaningful backlog.
Fix
Set SQS long polling to 20s in the queue and worker runtime without increasing task size.
Codify
Added a deploy-time guard plus an InfraProfiler rule so the same failure is ranked first next time.
Codex handoff
The report did not auto-mutate infra. It gave Codex the ranked target, telemetry evidence, owner repositories, and the exact guard to add after the fix was verified.
ExplicitMem Benchmark
Durable memory is the strongest proof point.
ExplicitMem turns agent memory into typed objects, graph links, bounded context synthesis, and receipt-backed retrieval. The local 500-question LongMemEval-S run proves retrieval quality; the hosted staging MCP route now has live latency smoke data from successful writes, searches, and context retrievals.
Accuracy
99.8%
Retrieval recall@K
100%
Hosted write p50
3.45s
Hosted search p50
3.28s
explicitmem / recall run
longmemeval-s-answer-v1 / 500 questions
ExplicitMem
Quality
99.8%
Latency
3.45s p50
Tokens
796tok
Mem0
Quality
94.4%
Latency
1090ms
Tokens
6787tok
Supermemory
Quality
85.4%
Latency
<300ms
Tokens
n/a
Hosted figures are from the staging API route on June 26, 2026: 10 successful samples, `project_specific` scope, bearer-authenticated MCP. Local evaluation latency remains a separate 30ms average benchmark artifact.
Operating Model
Each MCP should make its boundary visible before it touches the work.
01
Target
Define the repo, browser, memory scope, or release surface.
02
Run
Execute the MCP workflow through account-owned access.
03
Record
Keep screenshots, logs, traces, and notes attached.
04
Decide
Expand, change course, or stop with the evidence in view.
Access
Access opens when routes, account setup, and receipts are ready.
We are not publishing fixed pricing yet. Use the catalog to see what is coming and which MCPs already have research behind them.
Environment
Know which repo, browser surface, memory scope, or account the workflow will touch.
Evidence
Keep screenshots, logs, benchmark notes, and receipt files close to the work.
Handoff
Tie API access, billing, support, and rollback decisions to the same account.
What You Get
Hosted MCP products, account ownership, and enough evidence to know what happened.
The point is to make agent tooling less opaque: what the tool did, what evidence it produced, and where the boundary sits.
MCP access
A controlled surface for MCP tools, API keys, and target environment notes.
Owner record
The account owner, support path, and rollback decision stay attached.
Evidence pack
Screenshots, repository context, memory traces, logs, or receipt files stay available.
Exit criteria
A clear place to decide whether to expand, change course, or stop using a route.
Good Fit
Teams preparing autonomous engineering agents to touch real repos, browsers, releases, and customer workflows.
Join when you need durable memory, repository intelligence, browser evidence, and account-level control in one direction.
Bad Fit
One-off demos, unaudited automations, or local tools that do not need account-level control.
Monarchic is for engineering work that needs memory, evidence, boundaries, and review. Local experimentation should stay local.
Next Step