开发调试工具MCP服务器

Local MCP Server

MCP server for development and debugging tools — Atlassian (Jira, Confluence, Bitbucket), databases (PostgreSQL, Redis), messaging (Kafka), observability (Elasticsearch, Elastic APM, Loki, Tempo), design (Figma), and notes (Obsidian).

Each client group runs on its own port within a single container, so your agent can connect to individual clients independently.

Architecture

One container, multiple MCP servers running concurrently — one per client group, each on its own port:

Only ports for enabled clients are actually bound. Disabled clients don't start a server.

Setup

Copy the example env file and fill in your values:

cp .env.example .env

Environment variables

Feature flags — enable only the clients you need:

Atlassian Cloud (Jira + Confluence — one client, same site domain):

Bitbucket (separate client and credentials):

bitbucket_upload_image reads file_path inside the container. Copy images to ./uploads (mounted read-only at /uploads) and pass /uploads/<name>; then docker compose up -d to apply the mount.

Jira and Confluence typically share one Atlassian Cloud API token. Bitbucket uses its own credentials and is only initialized when ENABLE_BITBUCKET=true.

PostgreSQL supports multiple named hosts, each with its own credentials (e.g. microservices, merchant, openapipartner), each available per environment (dev, qa, uat, prod) — see PostgreSQL: multi-host support below.

Redis, Kafka, Elasticsearch, Loki, and Tempo tools support multiple environments (dev, qa, uat, prod) selectable per call via an environment parameter — see Multi-environment support below.

Figma:

Figma tools work without Dev Mode.

Obsidian:

Requires the Local REST API community plugin.

Elasticsearch / APM: configured per environment — see Multi-environment support.

Loki / Tempo: configured per environment. Their tools default to environment="all" and merge results from every configured environment for incident debugging.

PostgreSQL: multi-host support

PostgreSQL is modeled as a set of independently credentialed hosts (e.g. microservices, merchant, openapipartner), each available per environment (dev, qa, uat, prod). Every pg_* tool call requires host and environment; the MCP server resolves the matching connection string server-side using POSTGRES_URL_<HOST>_<ENV> env vars — credentials never leave the MCP server or get exposed to the calling agent.

Hosts are open-ended: adding a new host is just setting new POSTGRES_URL_<NEWHOST>_<ENV> vars — no code changes needed. Only host/environment combos with a non-empty URL are available; calling pg_* with an unconfigured combo returns a clear error listing which combos exist.

The optional db parameter on pg_* tools still lets the agent switch databases within the same host (default: the database from that host's connection URL).

Example tool call:

{ "tool": "pg_query", "arguments": { "sql": "SELECT 1", "host": "microservices", "environment": "dev" } }

Multi-environment support

redis_*, kafka_*, elasticsearch_*, and apm_* tools require an explicit environment. Loki and Tempo accept the same environment names and additionally default to all, which queries every configured backend concurrently and merges results. A failed environment is reported in the response without hiding successful results from other environments.

Configure connection settings per environment using suffixed variables. Only environments with a non-empty URL or bootstrap-server value are available; calling a tool with an unconfigured environment returns a clear error listing which environments are available.

Redis:

Kafka:

Elasticsearch / APM: (repeat suffix pattern per environment: _DEV, _QA, _UAT, _PROD)

Use either API key or username/password per environment, not both.

Loki / Tempo: (repeat suffix pattern per environment: _DEV, _QA, _UAT, _PROD)

Use either a token or username/password for a backend. A token takes precedence when both are set.

Aggregated Loki example:

{ "tool": "loki_search_logs", "arguments": { "query": "{service_name=\"checkout\"} |= \"error\"", "hours_ago": 2, "limit": 100 } }

Aggregated Tempo example:

{ "tool": "tempo_search_traces", "arguments": { "query": "{ status = error }", "hours_ago": 2, "limit": 20 } }

Example tool call targeting dev explicitly:

{ "tool": "elasticsearch_search", "arguments": { "index": "logs-*", "query": "error", "environment": "dev" } }

Kafka calls use the same explicit selection:

{ "tool": "kafka_list_topics", "arguments": { "environment": "uat" } }

Run locally

pip install -e .
python -m src.server

Run with Docker Compose

docker compose up -d         # start in background
docker compose down          # stop
docker compose up -d --build # rebuild and start

Run with Docker

docker build -t local-mcp .
docker run -d --rm --name local-mcp --env-file .env \
  -p 7373:7373 -p 7374:7374 -p 7375:7375 -p 7376:7376 \
  -p 7377:7377 -p 7378:7378 -p 7379:7379 -p 7380:7380 -p 7381:7381 \
  local-mcp

Tests

pip install -e ".[test]"
pytest

Tools

Atlassian — port 7373

PostgreSQL — port 7374

Redis — port 7375

Kafka — port 7376

Figma — port 7377

Obsidian — port 7378

Elasticsearch + APM — port 7379

Loki — port 7380

Tempo — port 7381