oracle/mcp

Letzte erfasste Quellaktivität: 9. Oktober 2026 um 17:04

SkillsMP-Katalog aktualisiert: 5. November 2026 um 19:57

gesammelte Skills: 2

GitHub-Stars: 471
GitHub-Forks: 155

Mit einem KI-Assistenten installieren

Diesen Prompt kopieren und in den verwendeten KI-Assistenten einfügen.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/oracle/mcp.

README des Repositorys

README-Quelldatei: README.md

README-Quelldatei lesen

Vollständige README anzeigen

# Oracle MCP Server Repository Repository containing reference implementations of MCP (Model Context Protocol) servers for managing and interacting with Oracle products. Each MCP server under `src/` may be written in a different programming language, demonstrating MCP’s language-agnostic approach. ## What is MCP? The Model Context Protocol (MCP) enables standardized, language-agnostic machine-to-machine workflows across data, models, and cloud resources. MCP servers implement specific tool suites, exposing them to MCP-compatible clients. ## Project Scope - **Proof-of-concept/Reference implementations:** This repository is not intended for production use; servers are provided as reference and for exploration, prototyping, and learning. - **Polyglot architecture:** Each `src/<server-name>/` directory represents a distinct MCP server, and these may use Python, Node.js, Java, or other languages. ## Prerequisites - Supported OS: Linux, macOS, or Windows (varies by server; check server README) - Git (for cloning this repository) - Internet access (for downloading dependencies) - *Cloud access*: Some servers require Oracle Cloud Infrastructure (OCI) credentials and configuration ([OCI docs](https://docs.oracle.com/en-us/iaas/Content/API/Concepts/sdkconfig.htm)) **Note:** Each MCP server has its own specific requirements (e.g., language runtime version, libraries). Always see the respective `src/<server>/README.md` for detailed setup instructions. ## Quick Start Follow these instructions to get started as quickly as possible. Once finished, look [here](#local-development) to set up your local development environment if you wish to [contribute](#contributing) changes. 1. Install `uv` from [here](https://docs.astral.sh/uv/getting-started/installation/) 2. Install python with `uv python install 3.13` 3. If you are using OCI servers, configure your [OCI authentication](#authentication) 4. Choose a server below and add it to your [MCP client configuration](#client-configuration) ### Choose an OCI server For most OCI users, start with [`oci-cloud-mcp-server`](src/oci-cloud-mcp-server/README.md). It uses the official OCI Python SDK directly, without OCI CLI subprocess calls, and is the recommended general-purpose entry point for OCI workflows. Use [`oci-api-mcp-server`](src/oci-api-mcp-server/README.md) instead when you specifically want an MCP server backed by the OCI CLI. It exposes tools for discovering and running OCI CLI commands. Choose one of the other purpose-built servers when you already know the Oracle product or OCI domain you want to work with. These servers target specific service and product workflows rather than providing a general OCI entry point. Browse the [`src/`](src/) directories and read the relevant `src/<server>/README.md` before configuring one. ### Recommended: OCI Cloud MCP Server Run the server over stdio: ```sh uvx oracle.oci-cloud-mcp-server@latest ``` Then add this minimal configuration to your MCP client. Replace `<profile_name>` with the OCI CLI profile configured during [authentication](#authentication). For macOS/Linux: ```json { "mcpServers": { "oracle-oci-cloud-mcp-server": { "command": "uvx", "args": [ "oracle.oci-cloud-mcp-server@latest" ], "env": { "OCI_CONFIG_PROFILE": "<profile_name>", "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` ### Alternative: OCI API MCP Server Use this CLI-backed option when you specifically need OCI CLI commands: ```sh uvx oracle.oci-api-mcp-server@latest ``` ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "command": "uvx", "args": [ "oracle.oci-api-mcp-server@latest" ], "env": { "OCI_CONFIG_PROFILE": "<profile_name>", "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` To connect to an OCI MCP server running in HTTP streaming mode: Assuming you started the server by running: ```bash ORACLE_MCP_HOST=127.0.0.1 ORACLE_MCP_PORT=8888 ORACLE_MCP_BASE_URL=http://127.0.0.1:8888 OCI_REGION=us-phoenix-1 IDCS_DOMAIN=<idcs_domain> IDCS_CLIENT_ID=<client_id> IDCS_CLIENT_SECRET=<client_secret> IDCS_AUDIENCE=<audience> uvx oracle.oci-cloud-mcp-server ``` Register `${ORACLE_MCP_BASE_URL}/auth/callback` as a redirect URI in the OCI IAM confidential application for the server. then place the following in your MCP client configuration: :warning: NOTE: the `type` attribute differs across MCP clients; some use `http` as the transport value while others (like Cline) expect `streamableHttp`. ```json { "mcpServers": { "oracle-oci-cloud-mcp-server": { "type": "streamableHttp", "url": "http://127.0.0.1:8888/mcp" } } } ``` `oracle.oci-api-mcp-server` is `stdio`-only. For OCI HTTP servers, `IDCS_REQUIRED_SCOPES` is optional; if unset, the default is `openid profile email oci_mcp.<server_name>.invoke`, where `<server_name>` is the package name without `oracle.oci-` and `-mcp-server`, with `-` replaced by `_`. For example, `oracle.oci-cloud-mcp-server` defaults to `openid profile email oci_mcp.cloud.invoke`. ## Running with podman Some MCP servers in this repository support running via podman. ### Installing podman Use the following instructions to install and run podman https://podman.io/docs/installation ### Building the Container Image You can build the container image using the following command. The command shows building the container image for the oci-api-mcp-server. ```sh moon run oci-api-mcp-server:containerize ``` The above command builds the container image tagged as `oracle.oci-api-mcp-server:latest`. ### MCP Client Configuration For examples of configuring MCP clients to run the server using podman, see the client-specific sections below. Configurations typically involve using `podman run` as the command, with appropriate flags and volume mounts for credentials if needed (e.g., mounting `~/.oci` for OCI servers running over `stdio` transport). Alternatively, if you want to use HTTP transport using the podman container, then start an OCI HTTP-capable MCP server using the following command and configure your client as mentioned in Quickstart section above. ```bash podman run -e ORACLE_MCP_HOST=0.0.0.0 -e ORACLE_MCP_PORT=8888 -e ORACLE_MCP_BASE_URL=http://127.0.0.1:8888 -e OCI_REGION=us-phoenix-1 -e IDCS_DOMAIN=<idcs_domain> -e IDCS_CLIENT_ID=<client_id> -e IDCS_CLIENT_SECRET=<client_secret> -e IDCS_AUDIENCE=<audience> -p 127.0.0.1:8888:8888 oracle.oci-cloud-mcp-server:latest ``` For local development, keep `-p 127.0.0.1:8888:8888`. Changing it to `-p 8888:8888` exposes the server beyond localhost. ## Authentication For OCI MCP servers running over stdio transport: 1. Install the [OCI CLI](https://docs.oracle.com/en-us/iaas/Content/API/SDKDocs/cliinstall.htm) 2. Configure your OCI CLI profile ```bash oci session authenticate --region=<region> --tenancy-name=<tenancy_name> ``` where: `<region>` is the region you would like to authenticate in (e.g. `us-phoenix-1`) `<tenancy_name>` is the name of your OCI tenancy Some MCP servers may not work with token-based authentication alone. See more about API key-based authentication [here](https://docs.oracle.com/en-us/iaas/Content/API/SDKDocs/clitoken.htm). All stdio actions are performed with the permissions of the configured OCI CLI profile. We advise least-privilege IAM setup, secure credential management, safe network practices, secure logging, and warn against exposing secrets. Remember to refresh the session once it expires with: ```bash oci session authenticate --profile-name <profile_name> --region <region> --auth security_token ``` `<profile_name>` is the profile that you set up in the steps above. You can view a list of your profiles by running `cat ~/.oci/config` on macOS/Linux if you forget which profile you have set up. For OCI MCP servers running over HTTP transport, use an OCI IAM confidential application and set `IDCS_DOMAIN`, `IDCS_CLIENT_ID`, `IDCS_CLIENT_SECRET`, `IDCS_AUDIENCE`, `ORACLE_MCP_BASE_URL`, `ORACLE_MCP_HOST`, `ORACLE_MCP_PORT`, and `OCI_REGION`. Register `${ORACLE_MCP_BASE_URL}/auth/callback` as a redirect URI in that application. HTTP requests run as the authenticated OCI IAM user and do not use the local OCI CLI profile for request authentication. `IDCS_REQUIRED_SCOPES` is optional; if unset, the server defaults to `openid profile email oci_mcp.<server_name>.invoke`. Create and grant that custom scope in your confidential application, or override it with `IDCS_REQUIRED_SCOPES`. For server authors, the shared library keeps credential resolution and HTTP token exchange consistent while each server retains its listener, service client lifecycle, and derived user agent. See the [shared authentication module](src/common/README.md#authentication-module) for the full configuration matrix and the [HTTP IDCS authentication section](src/common/README.md#http-idcs-authentication) for the provider and per-request token-exchange API. HTTP-derived OCI clients must be treated as caller-specific and must not be reused across callers. ## Client configuration Each MCP server exposes endpoints that your client can connect to. To enable this connection, just add the relevant server to your MCP client’s configuration file. You can find the list of servers under the `src` folder. Refer to the sections below for client-specific configuration instructions. ### Cline <details> <summary>Setup</summary> Before continuing, make sure you have already followed the steps above in the [Quick start](#quick-start) section. 1. If using Visual Studio Code, install the [Cline VS Code Extension](https://marketplace.visualstudio.com/items?itemName=saoudrizwan.claude-dev) (or equivalent extension for your preferred IDE). 2. Once installed, click the extension to open it. 3. Click the **MCP Servers** button near the top of the the extension's panel. 4. Select the **Installed** tab. 5. Click **Configure MCP Servers** to open the `cline_mcp_settings.json` file. 6. In the `cline_mcp_settings.json` file, add your desired MCP servers in the `mcpServers` object. Below is an example for for the generic OCI API MCP server. Make sure to save the file after editing. `<profile_name>` is the profile that you set up during the [authentication](#authentication) steps. For macOS/Linux: ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "type": "stdio", "command": "uvx", "args": [ "oracle.oci-api-mcp-server@latest" ], "env": { "OCI_CONFIG_PROFILE": "<profile_name>", "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` Alternatively, to run using podman (example for oracle.oci-api-mcp-server): ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "autoApprove": [], "disabled": false, "timeout": 60, "type": "stdio", "command": "podman", "args": ["run", "-i", "--rm", "-v", "/path/to/your/.oci:/app/.oci", "oracle.oci-api-mcp-server:latest"], "env": { "FASTMCP_LOG_LEVEL": "INFO" } } } } ``` Replace `"/path/to/your/.oci"` with the actual path to your OCI configuration directory. ⚠️ **NOTE**: Ensure that the _key_file_ field in /path/to/your/.oci/config uses the ~ character so that the path resolves both inside and outside the container; for example: `key_file=~/.oci/oci_api_key.pem`. For servers not requiring OCI credentials, omit the `-v` volume mount. For Windows - **TODO** 7. Once installed, you should see a list of your **MCP Servers** under the **Installed** tab. They will have a green toggle that shows that they are enabled. 8. Click **Done** when finished. </details> ### Cursor <details> <summary>Setup</summary> Before continuing, make sure you have already followed the steps above in the [Quick start](#quick-start) section. 1. You can place MCP configurations in two locations, depending on your use case: **Project Configuration**: For tools specific to a project, create a `.cursor/mcp.json` file in your project directory. This allows you to define MCP servers that are only available within that specific project. **Global Configuration**: For tools that you want to use across all projects, create a `~/.cursor/mcp.json` file in your home directory. This makes MCP servers available in all your Cursor workspaces. **`.cursor/mcp.json`** For macOS/Linux: ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "type": "stdio", "command": "uvx", "args": [ "oracle.oci-api-mcp-server" ], "env": { "OCI_CONFIG_PROFILE": "<profile_name>", "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` Alternatively, to run using podman (example for oracle-oci-api-mcp-server): ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "type": "stdio", "command": "podman", "args": ["run", "-i", "--rm", "-v", "/path/to/your/.oci:/app/.oci", "oracle.oci-api-mcp-server:latest"], "env": { "FASTMCP_LOG_LEVEL": "INFO" } } } } ``` Replace `"/path/to/your/.oci"` with the actual path to your OCI configuration directory. For servers not requiring OCI credentials, omit the `-v` volume mount. `<profile_name>` is the profile that you set up during the [authentication](#authentication) steps. For Windows - **TODO** 2. In your **Cursor Settings**, check your **Installed Servers** under the **MCP** tab to ensure that your `.cursor/mcp.json` was properly configured. </details> ### MCPHost <details> <summary>Setup</summary> Before continuing, make sure you have already followed the steps above in the [Quick start](#quick-start) section. 1. Download [Ollama](https://ollama.com/download) 2. Start the Ollama server For macOS: If installed via the official installer, `ollama start`. If installed via homebrew, `brew services start ollama` For Windows: If installed via the official installer, the server is typically configured to start automatically in the background and on system boot. For Linux: `sudo systemctl start ollama` 3. Verify the ollama server has started with `curl http://localhost:11434`. A successful response will typically be "Ollama is running". 4. Fetch the large language model, where `<model>` is the name of your desired model (e.g. `llama3.2`), with `ollama pull <model>`. For more options, check Ollama's list of [models that support tool calling](https://ollama.com/search?c=tools). 5. Install `go` from [here](https://go.dev/doc/install) 6. Install `mcphost` with `go install github.com/mark3labs/mcphost@latest` 7. Add go's bin to your PATH with `export PATH=$PATH:~/go/bin` 8. Create an mcphost configuration file (e.g. `~/.mcphost.json`). Check [here](https://github.com/mark3labs/mcphost?tab=readme-ov-file#mcp-servers) for more info. 9. Add your desired server to the `mcpServers` object. Below is an example for for the compute OCI MCP server. Make sure to save the file after editing. For macOS/Linux: ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "type": "stdio", "command": "uvx", "args": [ "oracle.oci-api-mcp-server" ], "env": { "OCI_CONFIG_PROFILE": "<profile_name>", "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` Alternatively, to run using podman (example for oracle-oci-api-mcp-server): ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "type": "stdio", "command": "podman", "args": ["run", "-i", "--rm", "-v", "/path/to/your/.oci:/app/.oci", "oracle.oci-api-mcp-server:latest"], "env": { "FASTMCP_LOG_LEVEL": "INFO" } } } } ``` Replace `"/path/to/your/.oci"` with the actual path to your OCI configuration directory. For servers not requiring OCI credentials, omit the `-v` volume mount. `<profile_name>` is the profile that you set up during the [authentication](#authentication) steps. For Windows - **TODO** 10. Start `mcphost` with `OCI_CONFIG_PROFILE=<profile> mcphost -m ollama:<model> --config <config-path>` 1. `<model>` is the model you chose above 2. `<profile>` is the name of the OCI CLI profile that you set up above 3. `<config-path>` is the path to the mcphost configuration json file that you made above </details> ## Local development On macOS, install proto with [Homebrew](https://formulae.brew.sh/formula/proto) using `brew install proto`, then run `proto setup` and open a new terminal. For other platforms, see the [proto installation guide](https://moonrepo.dev/docs/proto/install). From the repository root, run `proto install moon --immutable-lockfile` and `moon setup`. See [Running tasks with moon](#running-tasks-with-moon) for platform setup. For a Python server, run its tests with Moon and launch its source checkout directly with `uv`: ```sh moon run oci-api-mcp-server:test uv --directory src/oci-api-mcp-server run --locked oracle.oci-api-mcp-server ``` `uv run` creates and updates the server project's own `.venv` from its lockfile. The Common consumers resolve `oracle-mcp-common` to this repository's `src/common` through `[tool.uv.sources]` entries with `workspace = true`. Moon infers their project dependencies from these declarations so Common changes trigger consumer tests and lock checks. Build a distribution only when you need to check packaging, with `moon run oci-api-mcp-server:build`. To load this checkout in an MCP client, use the absolute path to the server project. For example: ```json { "mcpServers": { "oracle-oci-api-mcp-server": { "command": "uv", "args": [ "--directory", "<absolute path to this repo>/src/oci-api-mcp-server", "run", "--locked", "oracle.oci-api-mcp-server" ], "env": { "FASTMCP_LOG_LEVEL": "ERROR" } } } } ``` For other servers, use their project directory and the executable declared in their `pyproject.toml`. For Node.js, Java, or other runtimes, follow the server's README. ### JavaScript MCP servers Most servers in this repository are Python packages managed with `uv`. JavaScript servers are first-class MCP servers too, but they use `npm` and are intentionally excluded from the Python package loop. The OCI JavaScript MCP server lives in `src/oci-javascript-mcp-server`: ```sh moon run oci-javascript-mcp-server:test moon run oci-javascript-mcp-server:check moon run oci-javascript-mcp-server:build ``` ## Directory Structure ``` . ├── src/ │ ├── oci-api-mcp-server/ # MCP server (Python package) │ ├── oci-javascript-mcp-server/ # MCP server (Node.js package) │ ├── oracle-db-mcp-java-toolkit/ # MCP server (Java package) │ └── ... ├── LICENSE.txt ├── README.md ├── CONTRIBUTING.md └── SECURITY.md ``` Each server subdirectory includes its own `README.md` with language/runtime details, installation, and usage. ## Testing ### Testing with a Local Development MCP Server Use the [local development MCP client configuration](#local-development) above to launch the checkout. Changes to Python source are available on the next server start; no build or installation into a shared environment is needed. To test an HTTP-capable server locally: ```bash moon run oci-api-mcp-server:test ``` then start the server: ```bash ORACLE_MCP_HOST=127.0.0.1 ORACLE_MCP_PORT=8888 \ uv --directory src/oci-api-mcp-server run --locked oracle.oci-api-mcp-server ``` ### Inspector The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction) provides [Inspector](https://github.com/modelcontextprotocol/inspector) which is a developer tool for testing and debugging MCP servers. More information on Inspector can be found in the [documentation](https://modelcontextprotocol.io/docs/tools/inspector). The Inspector runs directly through npx without requiring installation. For instance, to inspect your locally developed server, you can run: ```sh npx @modelcontextprotocol/inspector \ uv \ --directory "<absolute path to this repo>/src/oci-api-mcp-server" \ run \ --locked \ oracle.oci-api-mcp-server ``` Inspector will run your server on localhost (for instance: http://127.0.0.1:6274) which should automatically open the tool for debugging and development. ### Running tests ```bash moon run :lint :lock-check :install-check :test :check :build moon run root:combine-coverage ``` CI uses `moon ci` with the same validation targets to run only tasks affected by the changed files and their dependencies. Pull requests compare the checked-out merge commit with the base commit; pushes compare with the previous commit. Python coverage is combined and uploaded only when Python tests produce coverage files. Changes to shared toolchain and CI configuration select all validation tasks. Project Moon configuration changes select that project's validation tasks. ### Running tasks with moon The standard Python server projects and the JavaScript MCP server are orchestrated with [moon](https://moonrepo.dev/). Tool versions are pinned in `.prototools`; after installing [proto](https://moonrepo.dev/docs/proto/install), install the pinned tools and run tasks from the repository root: ```bash proto install moon --immutable-lockfile moon setup moon run oci-compute-mcp-server:test moon run oci-javascript-mcp-server:test moon run root:lint moon run :build ``` The shared CI setup action downloads a pinned proto release and verifies its archive against the SHA-256 stored beside the version in the action before extracting or running it. It then follows the same sequence shown above. Moon's version and platform checksums are recorded in `.protolock`; proto verifies the Moon download against the lockfile. After changing the Moon pin, run `proto install moon` on Linux x64 and macOS ARM64 and commit the generated `.protolock` with `.prototools`. For another platform, run `proto install moon` once to generate its record before using `--immutable-lockfile`. Moon uses each project's language-specific package definition: `pyproject.toml` and `uv.lock` for Python, and `package.json` and `package-lock.json` for JavaScript. Build, test, and publish tasks are defined in Moon. Other projects excluded from Moon continue to use the validation commands in their own README. Use `uv lock` or `uv sync` from a Python project's directory when updating or syncing that project's dependencies. Optional end-to-end tests can be run with `moon run root:e2e-tests` after following the setup in `tests/README.md`. Python container images with a `Containerfile` can be built with `moon run <project>:containerize`. ## Publishing Use the manually dispatched [Publish package](.github/workflows/publish.yml) workflow from `main`. Select `pypi` or `testpypi` with a Python Moon project ID or `all-python` to release every Moon-managed Python package. Select `npm` with `oci-javascript-mcp-server` for the JavaScript package. Moon runs checks and build before the workflow passes the distributions to its publish job. Releasing a Python server publishes `oracle-mcp-common` first and waits for it to appear on the selected index. Ordinary CI never invokes the publish tasks. Configure the `pypi`, `testpypi`, and `npm` GitHub environments with release approval rules and their respective `PYPI_TOKEN`, `TEST_PYPI_TOKEN`, and `NPM_TOKEN` secrets before running the workflow. The publish job reads only the secret for its selected registry. The Moon publish tasks require the release workflow's `RELEASE_PUBLISH` flag, and Python publishing also requires an explicit upload and check URL. To verify a Test PyPI package after publishing: ```bash uvx --refresh-package oracle.oci-api-mcp-server \ --index https://test.pypi.org/simple \ --from 'oracle.oci-api-mcp-server==<published-version>' \ oracle.oci-api-mcp-server ``` Replace `<published-version>` with the version you released. ## Contributing This project welcomes contributions from the community. Before submitting a pull request, please [review our contribution guide](./CONTRIBUTING.md). ## Security Please consult the [security guide](./SECURITY.md) for our responsible security vulnerability disclosure process. ## License  Copyright (c) 2025 Oracle and/or its affiliates. Released under the Universal Permissive License v1.0 as shown at <https://oss.oracle.com/licenses/upl/>. ## Third-Party APIs Developers choosing to distribute a binary implementation of this project are responsible for obtaining and providing all required licenses and copyright notices for the third-party code used in order to ensure compliance with their respective open source licenses. ## Disclaimer Users are responsible for their local environment and credential safety. Different language model selections may yield different results and performance.

README des Repositorys wird geladen…

Skills in diesem Repository

Klassifizierung ausstehend

Skill

Beruf

Beschreibung

Aktualisiert

local-review

nicht klassifiziert

Use when asked to review local, committed, staged or uncommitted changes in this MCP repository, or perform a pre-PR review.

Quellsprache: Englisch

2026-10-09

pr-authoring

nicht klassifiziert

Use when asked to draft a PR title or description from a local branch in this MCP repository, fill its PR template, or create a draft PR.

Quellsprache: Englisch

2026-10-09

Es werden 2 von 2 gesammelten Skills angezeigt.