GitLab + Jira MCP Server
Enables read-only access to GitLab and Jira data through MCP, allowing users to list projects, merge requests, issues, and search Jira tickets via natural language queries. Provides safe, structured access to project management data without write permissions.
README
MCP GitLab + Jira (MVP)
This repository implements a read‑only Model Context Protocol (MCP) server that integrates with GitLab and Jira to expose project, merge request, and issue data to MCP‑enabled clients (e.g., editors or AI tools). It focuses on easy setup (env vars or a small JSON config), safety (no write actions in MVP), and a modular design that’s ready for future expansion.
How it’s built
- TypeScript with adapters for GitLab/Jira using a small HTTP helper (timeouts, retries, backoff) and structured logs with redaction.
- JSON‑RPC over stdio transport with tools listing and dispatch.
- Zod schemas validate inputs/outputs; errors are mapped via AppError into structured responses.
- Tests cover config, tool validation, adapters (mocked HTTP), and error sanitization.
Project history & planning
- Developer log: see developers_log.md for a step‑by‑step record of branches, changes, and merges.
- Task list: see task_list.md for current status and roadmap.
Architecture
High‑level components and request flow:
sequenceDiagram
participant C as MCP Client
participant S as MCP Server (CLI)
participant T as Tools/Handlers
participant G as GitLab API
participant J as Jira API
C->>S: JSON‑RPC (mcp.tools.call)
S->>T: Validate + dispatch (Zod)
alt GitLab tool
T->>G: REST (GET, paginated)
G-->>T: JSON data
else Jira tool
T->>J: REST (GET, paginated)
J-->>T: JSON data
end
T-->>S: Result (sanitized)
S-->>C: JSON‑RPC result
note over S,T: Logs are structured JSON with redaction
Module Map
src/cli.ts: CLI entry; parses flags, loads config, starts server; uses structured logging.src/mcp/jsonrpc.ts: Line‑delimited JSON‑RPC over stdio; parses requests and writes responses.src/mcp/server.ts: Wires adapters and tools; exposes tools.list and tools.call.src/tools/index.ts: Tool handlers; validates inputs/outputs with Zod; maps validation errors.src/adapters/gitlab.ts/src/adapters/jira.ts: Service clients with pagination and robust HTTP.src/lib/http.ts: requestJson with timeouts, retries, backoff; logs requests/responses with redaction.src/lib/errors.ts: AppError and toMcpError; sanitization.src/lib/log.ts: JSON structured logger with redaction.src/schemas/*.ts: Zod schemas for tool IO and schema metadata transformer.scripts/*: Demo JSONL requests and helper runner.tests/*: Unit tests for config, tools, adapters, and errors.
Error Propagation
- Validation: Zod throws; handlers wrap into
AppError('INVALID_INPUT', ...)→ transported as JSON‑RPC error data. - Config: Missing/invalid config raises
AppError('CONFIG_MISSING', ...)before server start. - Upstream: Adapters throw
AppError('UPSTREAM_HTTP_ERROR', ...)with status; HTTP helper retries on 429/5xx or timeouts. - Transport:
startJsonRpcStdiocatches errors, maps viatoMcpError, and returns{ error: { code: -32000, message, data } }. - Redaction:
sanitizeand logger redact tokens/headers and strip URL query params in logs.
Example JSON‑RPC error (shape):
{
"jsonrpc": "2.0",
"id": 7,
"error": {
"code": -32000,
"message": "Invalid input",
"data": { "code": "INVALID_INPUT", "message": "projectId is required" }
}
}
Install (dev)
- make install
- make build
- make run
Configure
- Env vars:
GITLAB_URL,GITLAB_TOKENJIRA_URL,JIRA_TOKEN
- Or JSON:
~/.mcp-gitlab-jira.json
{
"gitlab": { "url": "https://gitlab.com", "token": "<token>" },
"jira": { "url": "https://company.atlassian.net", "token": "<email:api_token>" }
}
Tools
- gitlab_list_projects
- gitlab_list_merge_requests
- gitlab_list_issues
- jira_search_issues
- jira_get_issue
Note: This MVP includes server/adapters stubs; transport wiring will evolve.
MCP Client Config
Add the server to your MCP-enabled client configuration (example):
{
"servers": {
"gitlab-jira": {
"command": "mcp-gitlab-jira"
}
}
}
Then invoke tools like gitlab_list_projects or jira_search_issues from your client.
Tool Examples
Assuming env is set and the server runs via stdio, these JSON-RPC lines illustrate usage.
Tools Overview
-
gitlab_list_projects: List projects accessible to the token. Example args:
{} -
gitlab_list_merge_requests: List MRs for a project; optional state. Example args:
{ "projectId": 123, "state": "opened" } -
gitlab_list_issues: List issues for a project. Example args:
{ "projectId": 123 } -
jira_search_issues: Search via JQL; supports maxResults. Example args:
{ "jql": "project=TEST", "maxResults": 10 } -
jira_get_issue: Get a Jira issue by key. Example args:
{ "key": "TEST-123" } -
List tools
- Input:
{ "jsonrpc": "2.0", "id": 1, "method": "mcp.tools.list" }
- Input:
-
GitLab: list projects
- Input:
{ "jsonrpc": "2.0", "id": 2, "method": "mcp.tools.call", "params": { "name": "gitlab_list_projects", "args": {} } }
- Input:
-
GitLab: list merge requests (opened)
- Input:
{ "jsonrpc": "2.0", "id": 3, "method": "mcp.tools.call", "params": { "name": "gitlab_list_merge_requests", "args": { "projectId": 123, "state": "opened" } } }
- Input:
-
GitLab: list issues
- Input:
{ "jsonrpc": "2.0", "id": 4, "method": "mcp.tools.call", "params": { "name": "gitlab_list_issues", "args": { "projectId": 123 } } }
- Input:
-
Jira: search issues (JQL)
- Input:
{ "jsonrpc": "2.0", "id": 5, "method": "mcp.tools.call", "params": { "name": "jira_search_issues", "args": { "jql": "project=TEST ORDER BY updated DESC", "maxResults": 10 } } }
- Input:
-
Jira: get issue
- Input:
{ "jsonrpc": "2.0", "id": 6, "method": "mcp.tools.call", "params": { "name": "jira_get_issue", "args": { "key": "TEST-123" } } }
- Input:
Makefile quick commands
# List tools
make list
# Call a tool
make rpc METHOD="mcp.tools.call" \
PARAMS='{"name":"gitlab_list_projects","args":{}}'
Demo Script
- Batch example requests are in
scripts/requests.jsonl(newline-delimited JSON). - Run them against the server:
make demo
Sample output (truncated):
{"jsonrpc":"2.0","id":1,"result":[
{"name":"gitlab_list_projects","params":{"type":"object","properties":{}},"result":{"type":"array"}},
{"name":"gitlab_list_merge_requests","params":{"type":"object","properties":{"projectId":{},"state":{}}},"result":{"type":"array"}}
]}
{"jsonrpc":"2.0","id":2,"result":[{"id":123,"name":"example"}]}
{"jsonrpc":"2.0","id":3,"result":{"issues":[{"key":"TEST-1"}]}}
Makefile
make install: Install dependenciesmake build: Compile TypeScript todist/make test: Run tests with coveragemake test-watch: Run tests in watch modemake lint/make lint-fix: Check/fix lint issuesmake format/make format-check: Apply/check formattingmake check: Run lint, format-check, and testsmake run: Start the server (uses compiled dist or tsx fallback)make list: Send mcp.tools.list via JSON-RPCmake rpc METHOD=... PARAMS='{}': Send arbitrary JSON-RPCmake demo: Build and run the demo JSONL requestsmake docker-build IMAGE=...: Build Docker imagemake docker-run IMAGE=...: Run Docker container (requires env vars)make add-remote NAME=gitlab REMOTE_URL=git@gitlab.com:group/repo.git: Add or update a git remotemake push-all REMOTE=gitlab: Push all branches to the specified remotemake push-tags REMOTE=gitlab: Push all tags to the specified remote
Logging
- Control verbosity with
LOG_LEVEL(debug,info,warn,error). Default:info. - Logs are structured JSON lines with redaction:
- Sensitive tokens and headers are replaced with
***. - URLs in HTTP logs omit query parameters.
- Sensitive tokens and headers are replaced with
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。