agentwebmap MCP server

Agent Web Map: an llms.txt directory for AI agents

Agent Web Map discovers and continuously validates websites publishing llms.txt. Search for a publisher or task, find original source links with matching evidence and freshness, and explore connections between the sections publishers include in their files.

People use the website. Agents use the same index through the free JSON API and hosted MCP server. Public discovery needs no account or API key. Connect your client directly to the hosted service; no local server installation is needed.

Find a source. Follow a connection. Bring it to your agent. Look up documentation for a task, discover related publishers, or save useful sources in your own workspace.

This repository is the public introduction and integration reference. The platform's application, crawler and infrastructure are maintained separately.

We build on the original llms.txt proposal by Jeremy Howard. Meet the llms.txt project, read its upstream source, and see how our independent directory uses the convention.

Start with a task

Try a real lookup on the homepage. Enter a publisher and task, inspect the original source links and actual API response, then read the publisher's pages. Match strength and stale flags make missing or older evidence visible. Refine weak results or use the publisher's search when coverage is insufficient.

For repeated use, connect MCP, use the public HTTP API, or explicitly install the optional discovery skill through your client's workflow. The skill describes when to try discovery, read primary sources and fall back. Connections make tools available; the host decides when to call them. An llms.txt or discovery manifest does not automatically install a connection or establish persistent agent adoption. First-use workflow.

The service combines task-focused source discovery, explicit freshness and provenance, a growing graph of shared publisher headings, and optional verified publisher controls. The workspace supports people who want to inspect and save sources; discovery remains available without signing in.

Start on the website

Goal	Open
Try a task and inspect the actual response	Live task lookup
Find a website or documentation for a task	Browse and search
Keep favorites and private project collections	Your workspace
Start with an editorial set of sources for a task	Curated collections
Explore shared publisher headings and connections	Interactive section cloud
See current counts, growth and file statistics	Statistics
Explore the index by inferred country	World map
Validate an existing llms.txt	Checker
Draft an llms.txt for a site	Generator
Submit a publisher's file for checking	Submit a site
Verify ownership and manage a listing	Owner dashboard

Search supports words and meaning, with task relevance before authority and ownership signals. Filter by category, publisher-declared MCP capability, kind of site, country, inferred action, or headings extracted from publisher files. Examples for agents.

Filter by country in Browse or the section cloud, or pass country=FR / country=unknown through REST and MCP. GET /v1/countries and list_countries expose available codes and listed-site counts. Country evidence can come from an owner declaration, a published organization address, a domain suffix or hosting. Responses distinguish the evidence; it does not establish verified headquarters or markets served.

For agent-readable orientation, llms.txt is the concise service guide. llms-full.txt includes the complete platform/API/MCP reference generated from the same documentation as docs.md, plus eligible owner-verified publishers. The guides cover country and section filters, the point cloud, public collections, optional private workspaces and owner management. Private account and admin records are excluded; unverified publishers remain discoverable through the public API.

The section catalogue grows as new files and headings are indexed. The interactive viewer opens topic details, matching publishers, filters and color layers when selected. Teal links show shared headings, violet links show inferred similarities in meaning, and optional amber highlights mark emerging headings observed in the index. Meaning connections do not change heading memberships or widen filters. Original headings and partial-coverage information remain available through the API. Section discovery guide.

Start from a website in Explore to inspect its headings and related publishers. Agents can use GET /v1/similar/{domain} or MCP similar_sites. Related sites are discovery suggestions; similarity does not imply a partnership or equivalent capabilities.

Public collections offer editorial starting points for building MCP tools, payments, and AI applications.

A workspace for people

The full-width directory interface offers light and dark modes, source previews, list and grid views, and a persistent sidebar. Sign in with an email code to save favorites, organize private collections and return to recently opened sources across browser sessions. Codes expire after ten minutes and work once in the requesting browser. Your source lists are private and separate from the public index. Copy a reference prompt, download sources.md, or connect your agent through MCP or REST.

Connect verified websites to see their owner dashboards from the workspace. Email sign-in confirms access to a mailbox; domain ownership still requires a fresh proof or the current owner key. Standalone owner-key sign-in remains available. Basic directory, API and MCP discovery stay anonymous. Free accounts unlock saved reports, named integrations, scoped API keys, generators, website audits, MCP endpoint checks and deeper enumeration. Live workspace reference.

Add it

MCP endpoint: https://agentwebmap.com/mcp — stateless Streamable HTTP, with no key required for public reads. Codex, Claude Code, Cursor, VS Code and other remote HTTP MCP clients can use the same endpoint.

Codex:

codex mcp add agentwebmap --url https://agentwebmap.com/mcp

Claude Code, for the current project:

claude mcp add --transport http agentwebmap --scope project https://agentwebmap.com/mcp

Cursor, in .cursor/mcp.json:

{"mcpServers":{"agentwebmap":{"url":"https://agentwebmap.com/mcp"}}}

VS Code, in .vscode/mcp.json:

{"servers":{"agentwebmap":{"type":"http","url":"https://agentwebmap.com/mcp"}}}

Merge the entry into an existing configuration. Connecting makes tools available; the client chooses when to call them. Try: “Use Agent Web Map to find the Model Context Protocol authorization documentation, then read the original source.”

Live MCP reference · Codex setup · Claude Code setup · Cursor setup · VS Code setup

Use the JSON API

curl --get https://agentwebmap.com/v1/resolve \
  --data-urlencode "q=modelcontextprotocol authorization" \
  --data-urlencode "limit=3"

Endpoint	Purpose
GET /v1	Discover API links and service information
GET /v1/resolve?q=modelcontextprotocol%20authorization	Find documentation URLs, evidence and freshness for a task
GET /v1/search?q=cloudflare	Search publishers by words and meaning
GET /v1/domains	Browse ranked listings with filters and pagination
GET /v1/countries	Available inferred country codes, names and listed-site counts, including unknown locations
GET /v1/domain/agentwebmap.com	Look up one publisher and its files
GET /v1/domain/agentwebmap.com/llms.txt	Read a validated cached publisher file
GET /v1/categories	Discover category IDs, counts and declared capabilities
GET /v1/sections	Discover heading labels, counts and emerging sections
GET /v1/section-graph	Explore heading connections and selected publishers
GET /v1/similar/cloudflare.com	Find related publishers, with optional country and action filters
GET /stats.json	Read current index statistics

API reference · Markdown reference · Live OpenAPI · Local reference

MCP tools

The hosted server exposes 25 tools, checked on 10 November 2026. tools/list is the live inventory.

Tool	Purpose
resolve	Find documentation URLs, evidence, freshness, terms and alternatives
search_llms_registry	Search publishers and indexed documentation metadata
search_index_preview	Return a small keyword preview without exhaustive counts or documentation hydration
list_domains	Browse and filter ranked listings
lookup_domain	Read a publisher's listing and file URLs
read_llms_txt	Read a cached file with source URL, check date and hash
list_categories	Discover categories, agentic labels and declared capabilities
list_countries	Discover inferred countries and counts; filter list, search, resolve and sections with country
list_sections	Discover headings, counts, emerging labels and coverage
explore_sections	Explore shared headings, connections and matching publishers
similar_sites	Find related publishers from a known domain
directory_stats	Read directory-wide counts
check_llms_txt	Validate a published file and see suggested fixes
validate_llms_txt	Check a pasted draft without fetching or retaining it
audit_website	Review dated public file, homepage and crawler-rule signals
check_mcp_endpoint	Inspect public MCP tool metadata without running tools
generate_llms_txt	Generate a draft for an owner to review before publishing
verification_instructions	Read ownership methods and how to start a private claim
register_domain	Submit a domain for checking
begin_ownership_claim	Start a fresh ownership challenge
verify_domain	Complete a challenge using private credentials and published proof
rotate_owner_key	Replace an owner's API key
update_site	Manage a verified listing and its terms for agents
site_stats	Read private health, discovery activity and score history
recrawl_site	Request a fresh check for a verified listing

Owner tools require the domain's owner credentials. Public discovery does not require ownership verification. Registration, ownership and listing updates should be performed at the owner's request. Ownership guide.

Why verify your website?

Verify your website—free to control its description and tags, publish terms for agents, recheck file health, and return to a private dashboard with daily discovery activity and score history. Preview the dashboard before verifying; the preview is illustrative and contains no live publisher data. Valid, listed, unblocked, verified profiles become eligible for search-engine visibility. Verified listings receive an ownership signal and an embeddable badge; eligible verified publishers are included in our llms-full.txt. Default Browse puts verified listings first; task search prioritizes relevance.

Unverified publisher profiles stay available on the website and through REST/MCP, with search indexing disabled. Search engines decide which eligible verified profiles to index; verification does not guarantee ranking or establish endorsement.

The browser guides you through proof in llms.txt, DNS, a homepage meta tag or a well-known file. API and MCP clients start a fresh claim, publish its proof and complete the check using private claim credentials. No backlink is required. Verification confirms domain control, rather than accuracy or endpoint availability.

Verified owners can edit the agent-facing summary in the management dashboard, REST PATCH /v1/domain/{domain} or MCP update_site, or restore the generated summary. The opening summary/Overview comes from the publisher’s llms.txt and changes through normal crawling. Listing health and activity are available in the owner dashboard. Public watches and outbound owner notifications have been retired. Current owner controls.

Access and source quality

Public discovery needs no account or API key and reads stored data only. It never fetches publisher websites or invokes paid AI. API pages hold at most 100 results; the website's Browse view defaults to 250 rows. Enumeration windows cover 250 anonymous or 1,000 registered results per query. Live website checks, generation, audits and MCP endpoint checks require free registration; paid-provider calls remain prohibited for registered clients too. MCP OAuth with PKCE and self-service integration keys are supported; domain-owner keys remain separate. Current access details.

Coverage is incomplete. Matching evidence comes from publisher metadata, link descriptions and headings, rather than the full text of every linked page. Read original sources before relying on implementation details. Use the publisher's search or web search when results are missing, weak or stale. Inferred labels and declared endpoints do not certify an available action. Categories and capabilities · Terms for agents.

Current figures are on the statistics page. Request counts can include crawlers, tests and repeated calls; they do not establish unique visitors or independent agent adoption.

Learn more and get in touch

• What is llms.txt?
• Real llms.txt examples
• llms.txt best practices
• How Agent Web Map works
• Agent-readable service overview
• Practical task examples
• Contact Agent Web Map · X

Feedback: Issues or Discussions.

The brand assets may be used to link to Agent Web Map. The MIT license covers the materials in this public repository; the hosted platform and directory data are provided separately by agentwebmap.com.

Public feature and interface information reviewed 10 November 2026. Use the live documentation and tools/list for the current service contract.