MCP Inspector

Inspecting PlanExe's MCP server.

This is my (Simon Strandgaard) preferred way to troubleshoot MCP. Whenever there is a problem, the MCP Inspector is the HAMMER.

Locations: Github, Documentation

Overview of PlanExe's MCP servers

PlanExe has multiple MCP servers that can be connected to.

#	Difficulty	Description
1	Beginner	MCP server at mcp.planexe.org/ and cost credits to use. Manage your credits via this page: home.planexe.org
2	Medium	MCP server inside docker on your own computer. I recommend using OpenRouter for inference, which cost money. You can be lucky finding a free model on OpenRouter, but this requires developer skills and several attempts. You can also run models on your own computer.
3	Expert	MCP server as a python program on your own computer.

Approach 1. MCP server at mcp.planexe.org

Purchase credits

1. Open home.planexe.org
2. Sign in with Google
3. Buy credits for 6 USD. It cost around 1.4 USD to create one plan, so that should be around 4 plans.
4. Click Generate new API key and copy it to clipboard. You will need this API key, in order to connect to the server.

Connect to MCP Server

npx @modelcontextprotocol/inspector --transport http --server-url https://mcp.planexe.org/mcp/

URL note: The canonical server URL is https://mcp.planexe.org/mcp (no trailing slash) — use this in JSON config files and registry entries. The trailing slash (/mcp/) is used above because the MCP Inspector CLI appends sub-paths to the base URL; without it the inspector may send requests to the wrong path. Both forms are accepted by the server.

This opens the inspector in a browser

In the left sidebar; Expand the Authentication section.

This is what the custom headers should look:

Do not use OAuth – PlanExe uses API keys, not OAuth. The OAuth flow will fail with "Failed to discover OAuth metadata".

1. Use Custom Headers instead: click + Add inside the Custom Headers section.
2. In Header Name, insert X-API-Key.
3. In Header Value, insert your API key (e.g. pex_...).
4. Click Connect.

If Connect fails with this error: "Connection Error - Check if your MCP server is running and proxy token is correct". This can happen if the Authentication section has incorrect data, so double check for typos.

If Connect fails with this error: Connection Failed: "TypeError: NetworkError when attempting to fetch resource.". This can happen if the Authentication section has incorrect data, so double check for typos.

If Connect still fails, then please report your issue on Discord.

When connected

When connected follow these steps:

1. In the topbar; Click on the Tools tab.
2. In the Tools panel; Click on the List Tools button.

Now there should be a list with tool names and descriptions:

example_plans
example_prompts
model_profiles
plan_create
plan_status
plan_stop
plan_retry
plan_file_info

When you inspect plan_create, the visible input schema includes prompt and optional model_profile.

Follow these steps:

1. In the Tools panel; Click on the example_plans tool to preview example output, or example_prompts to see prompt examples.
2. In the right sidepanel; Click on Run Tool.
3. The MCP server should respond with example plans or example prompts.
4. Optionally run model_profiles to inspect available model_profile choices before plan_create.

Approach 2. MCP server inside docker

Prerequisites

I assume you are able to create plans on your computer via the frontend_single_user web interface, http://localhost:7860/. It doesn't make sense proceeding if there is a problem with LLMs and no plans can be created.

Start docker

PlanExe's docker stack exposes the MCP endpoint on your loopback interface (default 127.0.0.1:8001/mcp/). Start with docker compose up and wait until you see mcp_cloud and /healthcheck like this:

mcp_cloud | INFO: 127.0.0.1:43988 - "GET /healthcheck HTTP/1.1" 200 OK

Start inspector

In a separate terminal; launch the inspector.

npx @modelcontextprotocol/inspector --transport http --server-url http://localhost:8001/mcp/

Once the UI opens in the browser, keep Authentication empty and click Connect.

When connected

Then open the Tools tab, click List Tools.

Click example_prompts, click Run Tool.

Approach 3. MCP server as a python program

If MCP had a built-in download mechanism, then there wouldn't be a need for this python program. As of 2026-Feb-12 MCP doesn't have such download mechanism, and developers make kludgy workarounds. The mcp_local/planexe_mcp_local.py proxy runs a tiny Python MCP server that forwards tool calls to the remote mcp_cloud while downloading reports into a local directory.

Prerequisites

I assume that you already have verified that things are working in "Approach 2. MCP server inside docker". If things are broken there, it makes no sense following the instructions here.

Start docker

PlanExe's docker stack exposes the MCP endpoint on your loopback interface (default 127.0.0.1:8001/mcp/). Start with docker compose up and wait until you see mcp_cloud and /healthcheck like this:

mcp_cloud | INFO: 127.0.0.1:43988 - "GET /healthcheck HTTP/1.1" 200 OK

Start inspector

In a separate terminal; launch the inspector.

On my computer, I launch the inspector like this:

You have to make these adjustments for your computer.

The PLANEXE_PATH is an absolute directory that PlanExe is allowed to write to. The downloaded files lands here.

The /absolute/path/to/PlanExe is where you have cloned the PlanExe repo.

npx @modelcontextprotocol/inspector \
  -e "PLANEXE_URL=http://localhost:8001/mcp" \
  -e "PLANEXE_PATH=/absolute/path/for/downloads" \
  --transport stdio \
  uv run --with mcp /absolute/path/to/PlanExe/mcp_local/planexe_mcp_local.py

Once the UI opens in the browser, click Connect.

When connected

Then open the Tools tab, click List Tools.

Click example_prompts, click Run Tool.