agama-redmine-mcp
MCP server for Redmine. Enables AI agents to work with issues using simple natural language commands.
Documentation
π¦ Agama β Redmine MCP Server
MCP server for Redmine. Enables AI agents to work with issues using simple natural language commands.
Works just as well as a personal tool for a single developer β or deployed once and shared by the
whole team.
ββββββββββββ MCP (stdio) βββββββββββββββββββββββ REST/JSON ββββββββββββ
β Agent β βββββββββββββββΆ β Redmine MCP Server β βββββββββββββΆ β Redmine β
β (client) β βββββββββββββββ β (this project) β βββββββββββββ β server β
ββββββββββββ tool results βββββββββββββββββββββββ HTTP ββββββββββββπ¬ Just ask
> "What issues are assigned to me? Sort them by priority."
> "Create a bug in the Web project: login button doesn't respond on mobile, mark it urgent."
> "Summarize issue #456 and all its subtasks."
> "How many hours did I log this week, broken down by project?"
Requirements
- Node.js 22+
- A Redmine instance with the REST API enabled (_Administration β Settings β API_)
- A Redmine API key β go to My account β API access key while signed in
Quick start
Most MCP clients (e.g. Claude Desktop) launch the server for you β configure it once and forget it:
{
"mcpServers": {
"redmine": {
"command": "npx",
"args": ["-y", "@0xkillaflow/agama-redmine-mcp"],
"env": {
"REDMINE_URL": "https://redmine.example.com",
"REDMINE_API_KEY": "your-redmine-api-key"
}
}
}
}For Claude Desktop, this goes in `claude_desktop_config.json` (**Settings β Developer β Edit
Config**). Restart the client and the Redmine tools will appear in its tool list.
To run it standalone from a shell instead:
REDMINE_URL=https://redmine.example.com REDMINE_API_KEY=your-redmine-api-key \
npx -y @0xkillaflow/agama-redmine-mcpThe server speaks MCP over stdio and logs to stderr, so stdout stays clean for the protocol.
Configuration
Set via environment variables, validated at startup. See `.env.example` for a
template.
| Variable | Required | Default | Purpose | |||
|---|---|---|---|---|---|---|
| `REDMINE_URL` | yes | β | Base URL of the Redmine instance | |||
| `REDMINE_API_KEY` | yes | β | API key for the acting user | |||
| `REDMINE_TIMEOUT_MS` | no | `30000` | Per-request timeout in milliseconds | |||
| `REDMINE_ALLOWED_DIRECTORIES` | no | _empty_ | Directories the attachment tools may use | |||
| `LOG_LEVEL` | no | `info` | `debug` \ | `info` \ | `warn` \ | `error` |
| `MCP_TRANSPORT` | no | `stdio` | `stdio` (only supported mode today) |
File access (`REDMINE_ALLOWED_DIRECTORIES`)
The attachment tools are the only ones that touch your filesystem, and they may
only touch it inside this allowlist β a `:`-separated list of absolute directories
(`;` on Windows). It governs both directions: reads for
`redmine_upload_attachment` and writes for `redmine_download_attachment`.
REDMINE_ALLOWED_DIRECTORIES=/Users/you/redmine-files:/tmp/redmine-downloadsIt defaults to empty, which means no local file access at all β both tools
refuse every path until you opt in. That default is deliberate: an agent that can
be talked into uploading a file is an agent that can be talked into uploading
`~/.ssh/id_rsa` or the `.env` holding your API key. Allowlist the narrowest
directory that does the job. Paths are fully resolved (`..` segments and symlinks
included) before the check, so a symlink inside an allowed directory cannot point
out of it, and downloads never overwrite an existing file.
Tools
Nineteen intent-shaped tools, each tagged with MCP safety annotations. Full reference:
| Tool | Kind | Description |
|---|---|---|
| `redmine_list_issues` | read | Search/filter issues by status, assignee, tracker, dates⦠|
| `redmine_get_issue` | read | Full issue detail, optionally with journals and attachments |
| `redmine_create_issue` | write | Create an issue |
| `redmine_update_issue` | write | Edit an issue's status, assignee, notes⦠|
| `redmine_delete_issue` | write Β· destructive | Permanently delete an issue β no undo, cascades |
| `redmine_manage_issue_watchers` | write | Add/remove a watcher (notification subscription) |
| `redmine_list_issue_relations` | read | Dependency links of an issue, or one relation by its id |
| `redmine_create_issue_relation` | write | Link two issues (blocks, precedes, duplicates, relatesβ¦) |
| `redmine_delete_issue_relation` | write Β· destructive | Remove one link β narrow in scope and reversible |
| `redmine_search` | read | Free-text search across issues, wiki, news, projects |
| `redmine_list_projects` | read | List visible projects |
| `redmine_get_project` | read | Full project detail, trackers/categories/activities |
| `redmine_list_time_entries` | read | Query logged time |
| `redmine_create_time_entry` | write | Log hours against an issue or project |
| `redmine_get_current_user` | read | "Who am I" β the API key's user, memberships, groups |
| `redmine_list_users` | read | Find users by name/status/group β admin-gated on most sites |
| `redmine_list_reference_data` | read | Statuses, trackers, priorities, activities, doc categories |
| `redmine_upload_attachment` | write | Upload a local file β token for an issue's `uploads` array |
| `redmine_download_attachment` | write | Save an attachment to a local directory (writes to disk) |
> Heads up: `redmine_delete_issue` is the tool that destroys data β Redmine has no trash, and
> the delete takes the issue's comments, time entries, attachments, and relations with it. It is
> annotated `destructiveHint: true` so an MCP client can require confirmation; if your client does
> not gate on that, treat every call as final. `redmine_delete_issue_relation` carries the same
> annotation but is far narrower: it removes one link between two issues, and
> `redmine_create_issue_relation` puts it back in a single call.
> Heads up: on write, Redmine silently ignores an unknown `tracker_id`, `status_id`, or
> `fixed_version_id` instead of erroring β always check the returned issue. The same applies to
> `is_private`, which is silently dropped if your role lacks the "set public/private" permission.
Security
The server runs locally and acts as a single user: your `REDMINE_API_KEY` never leaves your
machine except as a request header to your own Redmine instance, and the server can do nothing
your account couldn't already do. Multi-user HTTP transport is planned but not yet implemented β
selecting `MCP_TRANSPORT=http` exits with a clear error.
Local file access is off by default and confined to
`REDMINE_ALLOWED_DIRECTORIES` when you enable it. Nothing
else in the server reads or writes your filesystem.
Troubleshooting
| Symptom | Likely cause |
|---|---|
| Client shows no tools | Something besides JSON-RPC is writing to stdout β check any wrapper |
| Exits immediately on startup | A required env var is missing or invalid; the error names it |
| `401 Unauthorized` | API key missing/wrong, or REST API disabled on the Redmine side |
| `403 Forbidden` | The key's user lacks permission for that action |
| `404 Not Found` | The id doesn't exist or isn't visible to the key |
| `422 Unprocessable Entity` | Redmine rejected the write β check the field-level messages returned |
Documentation
- `docs/architecture.md` β design and request lifecycle
- `docs/tools.md` β generated per-tool reference
- `docs/adding-a-tool.md` β recipe for adding a new tool
- `CONTRIBUTING.md` β dev setup and contribution guide
Development
npm install # install dependencies
npm run dev # watch-run the stdio server
npm test # run the test suite
npm run build # bundle to dist/License
Frequently asked questions
What is agama-redmine-mcp?
agama-redmine-mcp is MCP server for Redmine. Enables AI agents to work with issues using simple natural language commands.
How do I install agama-redmine-mcp?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is agama-redmine-mcp open source?
Yes β it is hosted on GitHub at https://github.com/0xkillaflow/agama-redmine-mcp and has 20 stars.
Related MCP tools
The go-to web for your AI coding agent β local-first search, fetch, crawl & research over MCP. No API keys, no cloud, $0/query. Public beta.
Browser automation clicks buttons. OpenTabs calls APIs.
Open-source cross-agent memory layer for coding agents via MCP. Compatible with Claude Code, Codex, Cursor, Windsurf, Gemini CLI, Antigravity, OpenClaw, Hermes Agent, Oh-my-Pi, Pi, Copilot, Kiro, OpenCode, and Trae.
The Open-Source Multimodal AI Agent Stack: Connecting Cutting-Edge AI Models and Agent Infra
A MCP for Claude Desktop / Claude Code / Windsurf / Cursor to build n8n workflows for you
MCP server to provide Figma layout information to AI coding agents like Cursor
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP