Claude Chat MCP

A free, open-source collection of local connectors that link Claude Desktop to the self-hosted and enterprise tools your team actually runs.

Anthropic's built-in connectors mostly target the public cloud editions of tools (Confluence Cloud, GitHub.com, and so on). But many teams live on the self-hosted, Data Center, and on-prem editions running behind a company firewall, plus tools Claude does not support at all. This project fills that gap.

Each connector is a small, self-contained MCP server that Claude Desktop launches on your own machine. There is no cloud service, no hosting to set up, and it works behind corporate proxies. Your credentials stay on your computer and are only ever sent to your own server.

Open source under the MIT license. Built for Windows, macOS, and Linux.

Connectors

18 connectors, grouped by what they connect to. Each one is its own standalone server.

Atlassian

Connector	What it does
Confluence	Search and read self-hosted Confluence pages and spaces (Server / Data Center). Can create pages and comments in Writer mode.
Jira	Search and read self-hosted Jira issues and projects with JQL (Server / Data Center). Can create issues and comments in Writer mode.
Bitbucket	Browse self-hosted Bitbucket repositories, pull requests, and commits (Server / Data Center). Can comment on pull requests in Writer mode.

Dev

Connector	What it does
GitHub	Search and read GitHub Enterprise Server repos, issues, and pull requests. Can create issues and comments in Writer mode.
GitLab	Search and read self-hosted GitLab projects, issues, and merge requests. Can create issues and comments in Writer mode.
Jenkins	Browse self-hosted Jenkins jobs and builds. Can trigger builds in Writer mode.
Grafana	Browse self-hosted Grafana dashboards, data sources, and alerts. Can add annotations in Writer mode.
Sentry	Browse self-hosted Sentry projects and issues. Can resolve or ignore issues in Writer mode.

Productivity

Connector	What it does
Airtable	List bases, tables, and records, and read individual records. Can create and update records in Writer mode.
Redmine	Search and read self-hosted Redmine projects and issues. Can create issues and notes in Writer mode.
Mattermost	Read self-hosted Mattermost teams, channels, and messages. Can post messages in Writer mode.

Data

Connector	What it does
PostgreSQL	Read-only SQL: list databases, tables, and columns, and run guarded queries.
MySQL	Read-only SQL over MySQL (same tools as above).
MariaDB	Read-only SQL over MariaDB.
ClickHouse	Read-only SQL over ClickHouse (HTTP interface).
Oracle	Read-only SQL over Oracle. Requires a special build (see note below).
Elasticsearch	Search self-hosted Elasticsearch or OpenSearch indices and read mappings. Can index documents in Writer mode.
MongoDB	Query self-hosted MongoDB databases and collections. Can insert documents in Writer mode.

Oracle note: Oracle has no pure-Rust driver, so it is built behind an oracle cargo feature and needs Oracle Instant Client installed at runtime. The other seventeen connectors are fully self-contained with nothing extra to install.

Install

There are two ways to add a connector. Most people should use the control panel app.

Option A: The control panel app (recommended)

A small desktop app that does everything for you: it bundles the connectors, lets you fill in your details in a form, tests the connection, and writes the Claude Desktop config itself.

1. Go to the Releases page and download the installer for your operating system:
   • Windows: the .msi or -setup.exe file
   • macOS: the .dmg file
   • Linux: the .deb or .AppImage file
2. Run the installer and open Claude Chat MCP.
3. Pick a connector and click Configure.
4. Enter your details:
   • The link to your server (for example https://wiki.corp.com).
   • Your credentials (usually a personal access token or API key; username and password also work for some connectors).
   • Optionally a proxy or other network settings under Advanced (leave blank unless your IT team tells you otherwise).
5. Choose a permission role: Viewer or Writer (see Permissions below).
6. Click Test connection to confirm it works, then Install.
7. Fully quit Claude Desktop (on Windows, right-click the tray icon and choose Exit) and reopen it.

Now ask Claude things like "Search our Confluence for the deployment runbook" or "List the tables in the app database."

The app is also available in English and Vietnamese, and lets you update or remove connectors later from the same screen.

A note on installers: the installers are not code-signed yet. Windows SmartScreen or macOS Gatekeeper may warn you the first time you run them. This is expected for an unsigned open-source app.

WebView2 on Windows: the app uses Microsoft's WebView2 runtime, which ships with Windows 11 and recent Windows 10. If it is missing, the app will point you to the free download.

Option B: Direct .mcpb bundle (advanced)

If you would rather not install the control panel app, each connector is also published as a standalone .mcpb bundle that Claude Desktop can load directly.

1. On the Releases page, download the .mcpb file for the connector you want. Each bundle is universal and works on Windows, macOS, and Linux.
2. In Claude Desktop, open Settings > Extensions.
3. Double-click the .mcpb file, or drag it into that Extensions screen.
4. Fill in the connector's configuration fields (link, credentials, and so on).
5. Fully quit and reopen Claude Desktop.

Permissions

Every connector runs in one of two roles, which you choose at install time:

• Viewer (read only): Claude can read your data but cannot change anything. This is the safe default. Choose it unless you specifically need Claude to make changes.
• Writer (read and write): Claude can also create, update, and delete data through the connector, for example creating a Jira-style issue or updating an Airtable record.

The read-only Data connectors are always read only regardless of role. For the others, write tools are only enabled in Writer mode.

Privacy

• You supply your own credentials at install time. Nothing is bundled with the connectors and nothing is hard-coded.
• Connectors run entirely on your own machine and only talk to the server address you provide. No data is sent to this project or any third party.
• Credentials are stored by Claude Desktop in its own local config file, the same way any MCP server's settings are.

Build from source

For developers who want to build the connectors themselves.

This is a Rust workspace. You need the stable Rust toolchain.

cargo build --release          # build all connector binaries
cargo test                     # run the test suite

On Windows, the connectors link with MSVC, so you need the Visual Studio Build Tools (C++ workload and the Windows SDK). The repo ships a wrapper that loads the MSVC environment for you:

./scripts/cargo.ps1 build
./scripts/cargo.ps1 test

Each connector binary supports a couple of helper flags:

• --manifest prints the connector's config descriptor (the fields shown in the UI).
• --test-connection checks the credentials in the environment and exits.

The desktop control panel is a Tauri app that lives in crates/control-panel/. It is excluded from the main Cargo workspace and built separately with cargo tauri build. It embeds the connector binaries, so build those first. See the build steps in .github/workflows/control-panel.yml for the exact sequence the CI uses.

How it works

Claude Desktop launches one OS process per MCP entry, so this project ships one small server binary per connector. Each has its own process, its own tool namespace, and its own crash domain. A bug in one connector cannot take down another.

A shared connector-core crate handles the common plumbing (HTTP client, authentication, proxy and PAC support, retries), and server-runtime provides the stdio MCP host. Adding a connector means adding one small crate.

See ARCHITECTURE.md for the full design.

Contributing

Contributions are welcome, especially new connectors. The repo is built so that adding a connector is mostly copy, paste, and fill in. See CONTRIBUTING.md for the step-by-step guide.

License

MIT