atlassian-netscaler-mcp

An MCP (Model Context Protocol) server for Jira and Confluence Server / Data Center deployments that sit behind Citrix NetScaler SSO.

It captures the NetScaler session cookie via a real browser (Playwright), stores it in the OS keychain (or a 0600 file under $XDG_CONFIG_HOME/atlassian-mcp/cookies/ when no keychain is available), and attaches it to every Jira / Confluence request. API calls themselves authenticate with a Personal Access Token.

Why

Most Atlassian MCP servers assume Atlassian Cloud or a simple API token. Enterprise Server/DC installs typically need more:

• NetScaler / SSO in front of every endpoint
• Session cookies that expire on their own schedule
• Per-project custom fields that are required on create but not discoverable from the API
• Separate Jira and Confluence URLs, one shared gate

This server handles that and exposes a single MCP surface.

Features

• Unified Jira + Confluence tools in one server
• Single NetScaler login fronts both APIs — one browser window, one cookie
• OS keychain when available (macOS Keychain, libsecret, Windows Credential Manager); atomic 0600 file fallback otherwise, with a one-time warning
• Per-customer profiles for required custom fields and defaults
• SSE / HTTP transport via FastMCP
• Tolerant init handshake for clients that fire tool calls before the MCP initialize handshake completes

How auth works

┌──────────────┐    MCP/SSE    ┌──────────────────────┐   HTTPS + cookie + PAT   ┌──────────────┐
│  MCP client  │ ────────────► │   server.py (this)   │ ───────────────────────► │  Jira / Conf │
│              │ ◄──────────── │                      │ ◄─────────────────────── │   behind SSO │
└──────────────┘               └──────────┬───────────┘                          └──────────────┘
                                          │ on expired cookie
                                          ▼
                                  ┌───────────────┐   Playwright   ┌──────────────┐
                                  │   login.py    │ ─────────────► │  SSO browser │
                                  └───────┬───────┘                └──────────────┘
                                          │ store cookie
                                          ▼
                                OS keychain / 0600 file

The NetScaler cookie alone won't authenticate you to Jira/Confluence — it only gets you past the gate. The PAT you set in CONFLUENCE_TOKEN / JIRA_TOKEN is what the APIs actually check. Both are required.

Install

Requires Python 3.11+ and uv.

git clone https://github.com/paul-pfeiffer/atlassian-netscaler-mcp.git
cd atlassian-netscaler-mcp

# One-time: install the browser used for the SSO login flow
uv run --with playwright python -m playwright install chromium

Configure

cp .envrc.example .envrc
$EDITOR .envrc     # fill in URLs and tokens
direnv allow       # if you use direnv

Required:

ATLASSIAN_TOKEN is accepted as a fallback for both CONFLUENCE_TOKEN and JIRA_TOKEN if you have a single PAT that works for both.

Optional:

Run

./scripts/mcp-start

The first call without a valid cookie opens a Chromium window for SSO; complete the login and the server resumes.

Manual login

uv run login.py

One login covers both Jira and Confluence. The script exits as soon as you've reached the authenticated Atlassian view.

Wire into an MCP client

// e.g. Claude Desktop config
{
  "mcpServers": {
    "atlassian": {
      "url": "http://127.0.0.1:8000/sse"
    }
  }
}

Customer profiles

Some Jira instances require custom fields on issue creation that aren't discoverable from the API. Drop a profile under config/customers/<name>/profile.json — see config/customers/example/profile.json:

{
  "jira": {
    "project_overrides": {
      "MYPROJ": {
        "issue_type_overrides": {
          "Story": {
            "required_fields": ["customfield_12345"],
            "default_fields": {
              "customfield_12345": [{ "key": "DEFAULT-1" }]
            }
          }
        }
      }
    }
  }
}

Then set JIRA_CUSTOMER_PROFILE=<name>.

Profiles are validated against config/profile.schema.json at load time — a typo like default_fiedls fails loudly with a clear error instead of being silently ignored. Editors that honor the $schema field (VS Code, JetBrains) will autocomplete and live-validate while you edit.

Security notes

• The NetScaler cookie is the only secret stored on disk (or in the keychain). PATs come from env vars and are never persisted.
• File-fallback cookies are written atomically with mode 0600 under $XDG_CONFIG_HOME/atlassian-mcp/cookies/. The process logs a warning on first fallback use.
• Nothing in this repo sends data anywhere other than your configured Jira / Confluence URLs.

Roadmap

• A profile doctor CLI that previews which fields a given profile would inject for a project/issue-type
• Hot-reload profiles on file change

Contributing

Issues and PRs welcome. Don't commit .envrc or customer profile files — config/customers/*/ is gitignored except for example/.

License

MIT.

Use at your own risk. This software interacts with corporate auth systems and stores session cookies locally; make sure your use complies with your employer's policies and the terms of service of the systems you connect to.