Ask an AI assistant about your EPICS control system, whether you are on shift or writing code against it. Why is this PV disconnected? Which IOC was supposed to serve it? When did the archive last see a value, and did anybody log a problem with that device this week?
That is one question, and its answer sits in four different services. A facility runs its live layer, a channel registry, an archiver, an alarm logger and a logbook side by side. Each has its own port, its own query language, its own idea of how a date is written, and often its own login. This server puts them behind one set of Model Context Protocol (MCP) tools, live values over p4p and the other planes over their own REST APIs. The question is asked once instead of four times, and each service you have configured answers the part it holds.
Read-only by default. The one tool that can write to your control system is triple-gated, and off until you turn it on. Logbook writes sit behind a separate gate of their own. The full posture, including what leaves your machine, is Safety and network posture.
It also answers what no single service can. Which of the channels this IOC serves is nobody archiving? The registry holds one list, the archive holds another, and nobody holds the difference. An assistant asks for both and works it out, with no script to write first.
Project status. Pre-1.0 and under active development. Tools and APIs may still change between minor versions, so pin one if you depend on it.
I want to try it, and I have no control system. Install it (below), then follow the
quick start: three commands, and epics-testpv serves the PV
for you, so there is nothing else to obtain. No facility, no IOC, no EPICS Base, no ChannelFinder,
no archiver.
I want my coding agent to see the machine while I work. Take one of the ready-to-paste blocks from the MCP client integration guide and leave both write gates alone. Your agent can then read what a record really carries: its units, its limits, its enum labels, instead of you guessing them into the code. It still cannot write anything back.
I want to point it at my facility. The deployment guide is written for
that. It starts at epics-init, which prints a client-configuration block for one of four
deployment shapes and checks it in the same step, and it then walks the variables plane by plane,
the CA-bundle recipe for internal HTTPS, and the documented assumptions. Whichever way in, you
end at epics-doctor, which probes every configured plane read-only and tells you what your
instance actually reaches, and whether either write gate is armed and where it could write.
MCP is a small, standard protocol that lets an AI assistant (Claude, or any MCP client) call tools you expose to it. This server is such a tool provider for EPICS.
Its READ reach is decided by the launcher's EPICS search-path environment, not by this
server: the address lists, EPICS_PVA_NAME_SERVERS (TCP unicast, not subnet-bound), and the
auto-address search, which EPICS defaults to ON. Even an empty environment broadcasts PV
searches into the local subnets. The optional REST services stay off until their URLs are set.
PV write is the one reach the launcher does not decide: enabling it forces a loopback-only
search reach, and the process refuses to start otherwise.
Do not assume isolation from this document. Run epics-doctor to see what an instance actually
reaches, and read Safety and network posture.
An assistant does not have to ask: every read answer carries a reach field naming the plane
that served it and its scope, so no value can be quoted without both. The reach comes from the
configuration, not from a probe, and the field says so itself.
The server joins several planes of an EPICS installation, and everything it reports is framed in these terms. Full descriptions are in the tool reference.
| Plane | Source | Tools |
|---|---|---|
| Live | p4p (PVAccess; p4p 4.x offers no Channel Access) | get_pv_value, get_pvs, get_pv_info, monitor_pv, discover_pvs, set_pv_value |
| Registry | ChannelFinder | find_channels, list_channel_vocabulary, diagnose_connection, coverage_audit† |
| History | EPICS Archiver Appliance | is_archived, get_pv_history, get_archive_info, get_appliance_info, list_archived_pvs |
| Alarm | Phoebus Alarm Logger | is_alarm_configured, get_alarm_history |
| Naming | ESS Naming Service | lookup_device_name, diagnose_connection, crossplane_check† |
| Logbook | Phoebus Olog | search_logbook, get_log_entry, list_logbooks, list_tags, list_log_levels, create_log_entry, reply_to_log, update_log_entry, add_log_attachment, list_log_attachments, download_log_attachment |
| Display | .bob operator screens and .plt Data Browser trends (CS-Studio / Phoebus) |
validate_pvs†, crossplane_check†, coverage_audit†, find_device† |
| IOC | e3 st.cmd (+ optional .db) |
crossplane_check† |
† Needs the opi_navigation engine, which is not part of the published package, so a
core install does not register these four at all. Display and IOC are made of them alone
and have no tool there; Registry and Naming lose one each and keep the rest. Why the
engine is missing is in Related and roadmap.
The Live plane is the only authority for connected or disconnected. Every other plane is explanatory, and is withheld rather than reported as a false negative when its service is not configured.
- Python 3.12+
- p4p ≥ 4.2, installed automatically. It bundles the EPICS libraries, so no separate EPICS Base build is needed for the client.
- A reachable EPICS PV. Your control system once you widen the address list, an IOC of your own, or
the synthetic one
epics-testpvserves, which needs nothing beyond this package.
From PyPI, with uv (recommended):
uv tool install epics-mcpor with pip:
pip install epics-mcpFor the development version, install straight from the repository
(uv tool install git+https://github.com/epicDirk/EPICS-MCP); from a local checkout,
uv tool install . and pip install . do the same thing.
This installs the core server: live PV access, diagnosis, and the REST-service planes, plus seven
commands (epics-mcp, epics-init, epics-testpv, epics-doctor, epics-diagnose,
epics-crossplane, epics-coverage).
The last two are display-aware and need the opi_navigation engine, which is not part of the
package: they refuse with an explanation rather than doing their work. Four MCP tools need it
too and behave the other way round: the server does not register them at all, so a client
without the engine meets a shorter tool list rather than a refusal. The other five commands and
every non-display MCP tool work without it.
See Related and roadmap.
Check the commands are on your PATH before anything else, because the failure otherwise arrives much later and somewhere unhelpful:
epics-doctor --version # or: which epics-doctor / where.exe epics-doctorEvery command answers --help and --version with its own name and the installed version, the one
a bug report asks for, on a core-only install as well. If nothing is found, add the tool directory
your installer used to your PATH (uv tool update-shell does that for uv). A current Linux
distribution refuses a system-wide pip install, so use a virtual environment or
uv tool install. If the install ends in compiler output, p4p had no prebuilt wheel for your
platform and fell back to building from source, see Compatibility.
| Page | What it answers |
|---|---|
| Quick start | Try it in three commands, with no control system: the test PV, the configuration block, and the client restart |
| Tools, CLIs, resources and prompts | What can it actually do? Every tool by plane, the standalone CLIs, the resources and prompts |
| Configuration | Every EPICS_MCP_* variable, including TLS trust and the EPICS network block |
| Safety and network posture | What is gated, what is audited, what decides network reach |
| MCP client integration | Ready-to-paste .mcp.json blocks, where they go, and the restart that finishes the job |
| Deployment guide | Bringing it up in your facility, plane by plane, with epics-doctor |
| Troubleshooting | It did not work: symptom first, from "the client says nothing" to a rejected config file, and the running-server questions after it |
| Removing it again | Uninstall and downgrade, and the four things an uninstall leaves behind |
| Write-gate contract | What every in-server write gate must satisfy, and what a gate is deliberately not |
| Operating guide | The operational cookbook: service landscape, recipes, error signatures. An assistant reaches the same text through the get_guide tool, one named section at a time, and an application through the epics://guide resource |
| Architecture | The server → tools → services → clients layering and the plane model |
| Security policy | Reporting a vulnerability, and an honest statement of what the write gates are not |
| Contributing | Dev setup, the gate chain, Definition of Done |
| Known limits | What is deliberately not guarded, dated and measured, including the tempting repairs that were probed and rejected |
| Changelog | Release history |
Written and validated against a real installation at ESS: an e3 IOC over PVAccess, an Archiver Appliance, ChannelFinder, the Phoebus Alarm server and logger, an Olog logbook, and the ESS Naming Service. The quirks it compensates for were measured against those services, not read off a specification.
No site is hard-coded. Every service URL and network setting is an EPICS_MCP_* environment
variable, so deploying elsewhere means setting those variables, not changing code. Two defaults
carry a site-specific value (the ChannelFinder privacy allowlists), both documented as overridable
in the deployment guide.
Platforms. Installation is only as portable as p4p, which ships the EPICS libraries as
prebuilt wheels. Where a wheel exists, pip install just works; where it does not, pip falls back
to building from source, which needs a compiler and is not something this project tests.
| Platform | p4p wheel for Python 3.12+ |
Install verified |
|---|---|---|
| Linux x86_64 | yes | yes, in a clean container |
| Windows x86_64 | yes | yes, in a clean venv |
| macOS Apple Silicon | yes | not tested here |
| macOS Intel (x86_64) | yes, via the universal2 wheel (needs macOS 11+) |
not tested here |
| Linux aarch64 | no, source build | not tested here |
Services. Exercised against a local EPICS stack: an e3 test IOC (PVAccess), an EPICS Archiver
Appliance (single- or multi-instance), ChannelFinder, the Phoebus Alarm server and logger, and a
Phoebus Olog logbook. Archiver topology note: in a single-JVM appliance the MGMT and retrieval
webapps share a port, so leave EPICS_MCP_ARCHIVER_RETRIEVAL_URL empty; in a split deployment
MGMT (:17665) and retrieval (:17668) are separate ports. The one service plane absent from
that stack is the ESS Naming Service, for which this package ships no local stand-in, so
it is exercised against the facility instance alone.
No version of those services is listed here, and that is deliberate. The deployment guide explains why, and what to run to answer the question for your own installation.
The gate chain is uv-based: uv sync --extra dev --locked, then
uv run pytest and uv run pre-commit run --all-files. A full local install additionally needs
--group displays for the opi_navigation PV engine, and a fresh clone needs one further command
to wire the commit-message guard, which lives outside the tree and which nothing can install for
you.
Every one of those, what CI runs and what it deliberately cannot, the Definition of done and the commit style: CONTRIBUTING.md.
Four things: what exists, what is deliberately missing, what is only intended, and the answer that none of these pieces gives on its own.
What exists. Four tools (validate_pvs, crossplane_check, coverage_audit, find_device)
and two commands (epics-crossplane, epics-coverage) join live PVs with the display plane:
the macro-expanded PV inventory of .bob operator screens and of the .plt Data Browser trends
reached from them, read through the opi_navigation PV engine. Everything else in this package
installs and runs without any of it; the planes table above marks the four and says what each
plane keeps.
What is deliberately missing. Those six are not available in a published install, and cannot
be made available by any flag. opi_navigation lives in a private repository, so it is wired
into a dependency group rather than advertised as an extra. A group stays out of the published
metadata, where an unreachable dependency would be a promise nobody could keep, and a package
index rejects a direct git reference outright. The two commands refuse with an explanation instead
of failing obscurely; the four tools are never registered, so they are absent from the tool list
rather than present and refusing, and the server drops its own display claims with them.
What is intended, with no date attached. Two neighbouring pieces exist and are in daily use
here, both in private repositories: opi-live, a plugin that exposes a RUNNING CS-Studio, and
CS-Studio-MCP, an offline MCP server for the display files themselves. Opening the engine up is
tied to how those two hold up in practice. Neither those two nor the engine carries a release
commitment, and nothing on this page is a schedule. This page will say when one is published.
What the three answer together. This server knows that a PV is disconnected, newly in
alarm, or absent from the archive; it does not know what anyone has on screen at that moment.
CS-Studio-MCP knows which display file carries that PV and where on the file it sits, with
nothing open anywhere. opi-live knows which displays are open in a running CS-Studio right
now. So there is a question none of the three answers alone and all three answer together:
this PV just stopped updating, is it on a display somebody has in front of them, and where on
that display. Until the engine and those two neighbours are published, that answer stays
in-house.
MIT, see LICENSE.
Independently developed EPICS MCP server, originally seeded from Jacky1-Jiang/EPICS-MCP-Server (MIT) and since rewritten on FastMCP and the p4p library by epicDirk, with a write-safety layer, batch operations, PV monitoring, and, on an install that carries the display engine, cross-plane provenance and OPI validation.