What is an MCP server?
A Model Context Protocol server gives an AI assistant a defined set of external tools. Instead of asking the model to invent an itinerary from memory, a travel MCP can let it call a real catalog, pass structured choices such as city and pace, and receive a predictable result. MCP is the connection layer; the assistant still handles the conversation.
Remote MCP servers live on the public internet and are reached through one HTTPS endpoint. The official MCP specification calls the modern remote transport Streamable HTTP. CityWalk Plan uses that transport and is published in the official registry as com.citywalkplan/planner.
Is this really free on Claude and ChatGPT?
The server is free on both platforms; the client access is different. “Free MCP” can mean either that the server charges nothing or that the whole path, including the AI account, costs nothing. CityWalk Plan satisfies the first definition everywhere and the second definition on Claude Free today.
| AI plan | Paste a custom MCP URL? | What applies |
|---|---|---|
| Claude Free | Yes | One custom connector. This is the simplest completely free route. |
| Claude Pro / Max | Yes | Add remote custom connectors from Customize → Connectors. |
| Claude Team / Enterprise | Yes, with admin setup | An owner adds the connector for the organization; members then connect it. |
| ChatGPT Free / Go / Plus | Not as an arbitrary URL | Directory apps may be available, but direct custom-MCP creation is not currently documented for these plans. |
| ChatGPT Pro | Yes, for read-only MCPs | OpenAI says Pro users can connect MCPs with read/fetch permissions in developer mode. CityWalk Plan is read-only. |
| ChatGPT Business / Enterprise / Edu | Yes, when enabled | Full MCP is available in beta on ChatGPT web, controlled by admins and roles. |
These access rules are easy to conflate because OpenAI now uses “apps” and “plugins” around MCP-powered integrations. The reliable test is the product UI: if ChatGPT does not show Settings → Apps → Create for your account or workspace role, there is no user-side switch that forces it on. OpenAI also says directory app availability can vary by plan, workspace, region, role, and capability.
How to add CityWalk Plan MCP to Claude Free
Anthropic's current remote MCP guide says custom connectors are available on Claude Free, Pro, Max, Team, and Enterprise, with Free limited to one. The following steps work in Claude on the web; the connected service is then associated with your Claude account.
- Open Connectors. In Claude, go to Customize → Connectors.
- Create a custom connector. Click the + button next to Connectors, choose Add custom connector, and name it CityWalk Plan.
- Paste the endpoint. Enter
https://citywalkplan.com/api/mcp. Leave OAuth client ID and secret empty; CityWalk Plan does not use login or authentication. - Add and enable it. Finish the setup, open a new conversation, click the + menu, then enable CityWalk Plan under Connectors.
- Run a clear first prompt. Ask: “Use CityWalk Plan to make an easy two-day walking itinerary for Lisbon.”
Claude should discover three tools. If it replies from general knowledge without calling one, mention the connector by name and say “use the plan_city_walk tool.” Once the first call works, normal travel questions are usually enough.
How to add the MCP server to ChatGPT
As of July 29, 2026, OpenAI's current developer-mode documentation says ChatGPT Pro can connect MCPs with read/fetch permissions, while full MCP is available in beta for Business, Enterprise, and Edu on the web. CityWalk Plan has no write actions, so it fits the read-only Pro boundary. Managed-workspace admin and role settings still apply.
- Turn on developer mode. On Pro, open Settings → Apps → Advanced Settings. In a managed workspace, an admin or owner enables developer mode or grants the required role.
- Open the app creator. In ChatGPT on the web, go to Settings → Apps → Create. If Create is absent, your current plan or role does not expose this path.
- Enter the server details. Use CityWalk Plan as the name, paste
https://citywalkplan.com/api/mcp, choose No authentication, then scan the tools. - Create and test the app. Start a new chat, select the draft or enabled CityWalk Plan app from the tools menu, and use the same Lisbon prompt.
What the existing CityWalk Plan MCP actually does
This is not a mock endpoint made for the tutorial. The production server was initialized and its registry status was checked again on July 29, 2026. It exposes three small, focused tools:
| Tool | Use it for | Main inputs |
|---|---|---|
list_cities | See the 50+ supported destinations, grouped by region. | None |
plan_city_walk | Build a 1–7 day itinerary from real catalog places, grouped geographically by day. | City, days, easy / balanced / packed pace, optional must-sees, language |
get_place | Look up one catalog place and return its area, visit time, coordinates, description, and source link. | City and place name |
A real result from the MCP server
For the prompt “Use CityWalk Plan to make an easy two-day walking itinerary for Lisbon”, the live planner returned six catalog stops. The assistant can explain them conversationally, but the underlying place selection comes from CityWalk Plan:
Day 2: Tram 28 → São Jorge Castle → Alfama
The point of the MCP connection is not that an LLM can write another travel paragraph. It is that Claude or ChatGPT can request structured, real places from a purpose-built planner, then hand the result back in the conversation where you can ask follow-up questions.
Prompts that reliably trigger the tools
Start with the connector name, a supported city, a day count, and a pace. Add constraints only when they are useful.
- “Use CityWalk Plan to make a balanced three-day walking itinerary for Kyoto. Include Fushimi Inari if it is in the catalog.”
- “Ask CityWalk Plan which European cities it supports, then recommend three for a first solo city walk.”
- “Use CityWalk Plan for one packed day in Rome. Return the stops by day and include the editable plan link.”
- “Look up Belém Tower with CityWalk Plan, then explain where it fits in a Lisbon walking day.”
Do not ask the MCP server for live museum hours, restaurant availability, hotel prices, weather, or street-by-street navigation; those are outside its current catalog. You can ask the AI assistant to research those separately, but it should cite current official sources and keep that information distinct from the CityWalk result.
Troubleshooting
“Add custom connector” or “Create” is missing
On Claude Free, check Customize → Connectors and remember that the plan allows one custom connector. Remove an old custom connector if you have already used the slot. On ChatGPT, a missing Create option normally means the current plan, workspace, or role does not allow direct custom MCP creation.
The tool scan fails
Use the exact HTTPS endpoint https://citywalkplan.com/api/mcp, select no authentication, and retry. Do not paste the human guide URL https://citywalkplan.com/mcp; that page explains the service but is not the JSON-RPC endpoint.
The assistant answers without using CityWalk Plan
Enable the connector for that conversation and be explicit once: “Use the CityWalk Plan connector and call plan_city_walk.” Starting a new chat after adding a connector can also clear stale tool state.
The city is not recognized
Ask the assistant to call list_cities first. The server intentionally rejects unsupported destinations instead of inventing a route. That is a feature: a clear “not in the catalog” is safer than a confident fictional result.
The order needs adjustment
Open the returned CityWalk Plan link and edit it on the map. Reorder or remove stops, then share or export the result. The draft groups places geographically, but your hotel, tickets, mobility needs, and current street conditions may justify a different order.
Privacy and safety
CityWalk Plan's MCP is public, read-only, and requires no CityWalk account. It does not ask for access to email, files, contacts, payment details, or another service. The assistant still sends the tool arguments needed for the request—such as city, days, pace, must-see place names, and language—to the endpoint.
Only add remote MCP servers from organizations you trust. Both OpenAI and Anthropic warn that third-party MCP servers can receive data from tool calls and may change their behavior. Never paste passwords or private booking records into a public travel-planning tool that does not need them.
Frequently asked questions
Can I use a custom MCP server on Claude Free?
Yes. Anthropic says Claude Free users can add one custom connector that uses a public remote MCP server. CityWalk Plan fits that route because its endpoint is public and needs no authentication.
Can I add an arbitrary MCP server to ChatGPT Free?
Not through the current direct custom-MCP creation flow. OpenAI currently documents read-only MCP connections for Pro and full MCP for eligible Business, Enterprise, and Edu workspaces. Free users can use directory apps when those apps and capabilities are available to their plan, but an MCP Registry listing is not automatically a ChatGPT directory app.
Is the CityWalk Plan MCP server itself free?
Yes. The hosted endpoint is free, public, and requires no CityWalk Plan account, API key, OAuth flow, or payment. Your AI assistant's own plan and usage limits still apply.
What URL should I paste?
Use https://citywalkplan.com/api/mcp. It is the remote Streamable HTTP endpoint. The shorter /mcp page is the human setup guide, not the server endpoint.
Does the server invent attractions with AI?
No. Routes come from a fixed catalog of real, hand-curated places with map coordinates. The connected assistant presents and discusses the result, but it does not create the catalog.
Does it provide live street routing and opening hours?
No. It groups real places geographically and orders them into a practical draft, but it does not claim live street routing, current opening hours, weather, or verified accessibility. Check current details before travel.
Official sources and verification
- Anthropic: Get started with custom connectors using remote MCP — plan availability, one-connector Free limit, setup, and security.
- OpenAI: Developer mode and MCP apps in ChatGPT — current Pro read-only access, full-MCP availability, and workspace setup.
- OpenAI: Apps in ChatGPT — directory discovery, plan and workspace variability, and app permissions.
- Model Context Protocol: Transports — Streamable HTTP behavior.
- Official MCP Registry record for CityWalk Plan — active listing, endpoint, and server metadata.