MCP server
Connect AI agents to the local food network
Food Web runs a remote Model Context Protocol server at https://mcp.foodweb.network. Claude, Cursor and any MCP client can search Canada's community-maintained map of kitchens, farms, markets and food banks, file and review map suggestions, and take part in the community forum — as your account, with your roles, audited on every write.
Fact sheet
What an agent can do
Query the asset map
Free-text and geographic search over food infrastructure — commercial kitchens, farms, markets, food banks, restaurants — with addresses, coordinates, contact details, tags, and links to bookable listings. Anonymous. Paginated. No key required.
Curate the data
Contributors file suggestions with duplicate detection; curators review queues, edit entities and places, merge duplicates, enrich from Google Places, and re-ingest from source websites — the full moderation pipeline, exposed as tools.
Participate in the forum
Read public discussions anonymously; with a key, start threads, reply and vote in the Food Web community forum as your account — visibility rules (public / members / private) enforced server-side.
Connected in one command
Works with Claude Code, Claude Desktop (via mcp-remote), Cursor, and any client that speaks MCP over HTTP. Mint a free fw_ key on the dashboard, or skip the key entirely for public reads. Then ask your assistant to run whoami.
claude mcp add --transport http foodweb https://mcp.foodweb.network \
--header "Authorization: Bearer fw_YOUR_KEY"Frequently asked questions
What is the Food Web MCP server?
It is a remote Model Context Protocol (MCP) server at https://mcp.foodweb.network that lets AI assistants like Claude, Cursor and other MCP clients call typed tools against the Food Web platform: searching the food asset map (kitchens, farms, markets, food banks and more across Canada), submitting and reviewing map suggestions, and reading or posting in the community forum on behalf of your account.
What can an AI agent do with it?
Anonymous agents can search mapped food infrastructure and read public forum threads. With a free fw_ API key the agent also posts and votes in the forum as your account. Accounts with the contributor role can file map suggestions with built-in duplicate detection, and curators/admins get the full curation surface: reviewing suggestions, editing entities and places, merging duplicates, archiving, Google Places enrichment, and website re-ingestion.
How do I connect Claude or Cursor to Food Web?
Add the endpoint https://mcp.foodweb.network as a remote MCP server. In Claude Code: claude mcp add --transport http foodweb https://mcp.foodweb.network --header "Authorization: Bearer fw_YOUR_KEY". In Cursor, add the same URL and header to .cursor/mcp.json. Claude Desktop connects through the mcp-remote bridge. Pre-filled snippets are on the developer dashboard.
Is it free? Do I need an account?
Yes — the MCP server is free, and public read tools work with no account at all. Sign up at api.foodweb.network to mint a free fw_ API key when you want the agent to act as you (forum posting, map suggestions, curation if you hold those roles).
What data does the asset map contain?
Community-maintained records of food infrastructure: commercial and community kitchens, farms, markets, restaurants, cafés, food banks and related assets — each with a geocoded place (address and coordinates), contact details, tags, and links into Food Web apps such as bookable kitchen listings.
Is it safe to give an AI write access?
Writes are scoped by your real roles, read live from Firebase custom claims (revocation propagates in about a minute), and every mutating tool call is recorded in an audit log with the actor, key, tool, arguments and outcome. Keys are stored as SHA-256 hashes and can be revoked at any time from the dashboard.
Which protocol and transport does it use?
MCP over Streamable HTTP in stateless mode: each JSON-RPC message is a single POST with a JSON response, so it works reliably from serverless clients and proxies. The server performs no AI inference — the model always runs client-side.
More on how Food Web works with AI: foodweb.network/ai