openllmc
Unified MCP server exposing the full OpenLLM gateway API (one tool per operation), semantic code/docs search, and cross-session memory.
README
<p align="center"> <picture> <source media="(prefers-color-scheme: light)" srcset="./assets/openllm-light.svg"> <img alt="OpenLLM" src="./assets/openllm.svg" width="300"> </picture> </p>
<p align="center"><b>openllmc</b> — the OpenLLM CLI.</p>
<p align="center"> <a href="./LICENSE"><img alt="License: BUSL-1.1" src="https://img.shields.io/badge/license-BUSL--1.1-blue.svg"></a> <img alt="source-available" src="https://img.shields.io/badge/source-available-informational.svg"> <img alt="targets" src="https://img.shields.io/badge/targets-darwin%20%C2%B7%20linux%20(arm64%2Fx64)-lightgrey.svg"> </p>
One compiled binary that carries every OpenLLM extension: a single MCP
server (openllmc mcp) exposing the full native gateway API (one tool
per OpenAPI operation, generated from the same schema the gateway serves at
/api/swagger), semantic code + docs search (claude-context), and
persistent cross-session memory (supermemory) — plus the hook
subcommands the openllm plugin's session hooks shell out to.
Self-contained by construction: the SDK is generated and committed
(src/sdk/generated/) so this package links no
private workspace code — it builds standalone from this repo.
Install
The canonical distribution is the compiled binary (verified against its published SHA-256). Installing the openllm plugin from your gateway's dashboard installs it for you; standalone:
curl -fsSL "https://openllm.sh/api/setup/cli/install.sh" | bash
# ↑ also runs `openllmc setup` (PATH symlink + shell completion)
openllmc version
openllmc mcp # the unified MCP server (stdio)
Installed from the dashboard (one-click)? That path runs sandboxed and skips PATH/completion — run
~/.openllm/bin/openllmc setuponce.
Or consume the source as a package:
bun install github:quantidexyz/openllmc # latest
Build from source
This repo is self-compilable — no private packages, no monorepo needed
(the SDK is committed under src/sdk/generated/):
git clone https://github.com/quantidexyz/openllmc && cd openllmc
bun install
bun run compile:host # → dist/openllmc (this machine's target)
./dist/openllmc version # v0.0.0-dev — source builds carry the dev sentinel
bun run compile # all 4 targets: darwin/linux × arm64/x64 (+ .gz sidecars)
A source build bakes the 0.0.0-dev sentinel (release builds are stamped by
the release pipeline via --version), so it never self-updates — run it
directly or point mcpServers.openllm at your dist/openllmc.
Commands
| Command | What |
|---|---|
openllmc mcp [--only <group>] |
the unified MCP server over stdio — groups: openllm (native API), claude-context, supermemory (default all) |
openllmc exec ctx <index|search|status|index-docs> … |
code/docs-search hook verbs (what the plugin's hooks call) |
openllmc setup |
PATH symlink + shell completion (idempotent) |
openllmc completion <bash|zsh|fish|install> |
print or install shell completion |
openllmc api --spec |
print the embedded OpenAPI spec |
openllmc self-update |
converge to the gateway's pinned release |
openllmc version |
print the version |
Config: LLM_GATEWAY_URL / LLM_GATEWAY_API_KEY (env, or
the shared ~/.openllm/.env — the same file the
daemon pairing writes, so OPENLLM_CLOUD_ORIGIN / OPENLLM_API_KEY apply to
every OpenLLM tool on the box).
Verify
Every published binary is pinned by SHA-256 in manifest.ts,
committed to this repo. Confirm the artifacts the cloud serves are exactly what
this source vouches for — no trust required:
bun install
bun run verify # download every published target, hash it, check vs manifest.ts
bun run verify -- --host # just this machine's target
bun run verify -- --file ./openllmc # a binary you already installed/downloaded
bun run verify -- --installed # the `openllmc` on your $PATH
Exit code is 0 only when every checked binary matches its pinned digest.
Note: the binary is not byte-reproducible (
bun build --compile --bytecodeembeds non-deterministic bytecode), so rebuilding from source won't hash-match the release. The verifiable guarantee is that the published asset matches the SHA-256 committed here — the same digest the install script andopenllmc self-updateenforce on download.
License
Source-available under the Business Source License 1.1 (© Quantide LLC) — converts to MIT on the Change Date. Not OSI open-source.
Read-only mirror. Regenerated from the OpenLLM monorepo each release. PRs welcome — ingested upstream with your authorship preserved. BUSL contributions require the CLA (the bot will prompt you).
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。