directree's Public MCP Server for AI Assistants

AI assistants can now query directree's software directory directly. Today we shipped a public MCP server and a small REST API that give agents structured access to tool listings, categories and comparison facts.

This is an early step, not a claim that directree will appear in every AI answer. It is a practical way to make the directory easier for assistants to read, query and cite when a user is looking for software.

What we shipped

The new directree MCP server is available at https://www.directree.io/api/mcp. MCP, or Model Context Protocol, is an open standard for connecting AI assistants to external tools and data sources. Instead of an assistant having to guess how to call a service, an MCP server describes the available tools and their inputs in a consistent format.

Our server currently exposes four tools:

• search_tools finds software by query, category, pricing and other filters.
• get_tool returns the structured facts for one listing.
• compare_tools puts multiple listings side by side.
• list_categories returns the categories available in the directory.

We also shipped an agent-readable REST API:

• GET /api/agent/tools for search and discovery
• GET /api/agent/tools/[slug] for one tool's facts
• GET /api/agent/categories for the category list

The same endpoints are public at https://www.directree.io.

Why this matters if your tool is listed

A listing is useful when a person can find it. It is also useful when a machine can understand it.

When an AI assistant receives a question such as "What are some simple analytics tools for a small SaaS?", it needs more than a pile of marketing pages. It needs to identify products, understand their categories, compare relevant facts and decide which sources are worth citing. A structured directory gives it a place to do that work.

If your tool has a listing on directree, an assistant can now search for it through the MCP server, retrieve its facts and use directree as one source in an answer. That does not guarantee a recommendation. It does mean your listing is available through a direct query path instead of depending only on a crawler discovering it at the right moment.

This is part of a broader shift in software discovery. The how AI search engines pick which tools to recommend post explains why clear entities and independent sources matter. The new API is one more way to make those facts accessible.

How to use it

Connect the MCP server to Claude Desktop

Add the server to Claude Desktop's configuration. The exact location of the file depends on your operating system, but the JSON looks like this:

{
  "mcpServers": {
    "directree": {
      "url": "https://www.directree.io/api/mcp"
    }
  }
}

Restart Claude Desktop after saving the configuration. You can then ask it to search directree for tools, inspect a listing or compare products. The directree MCP server documentation has the current tool definitions and connection details.

MCP clients and configuration formats are still changing, so check the documentation for client-specific instructions rather than assuming every assistant handles remote servers in exactly the same way.

Query the REST API

For a direct HTTP request, search the agent endpoint with a query parameter:

curl "https://www.directree.io/api/agent/tools?q=analytics&limit=5"

To retrieve a single listing, use its slug:

curl "https://www.directree.io/api/agent/tools/plausible"

The response is intended for machine use, but it is also readable enough to inspect with curl while building an integration. Categories are available here:

curl "https://www.directree.io/api/agent/categories"

Why provenance labels matter to agents

Machine-readable data is only useful if an agent can tell how trustworthy each field is. directree does not present every sentence as equally certain.

Our listings distinguish between three provenance levels:

• Observed means a fact was found or verified on the public site, such as a visible pricing page, integration or platform.
• AI-inferred means directree generated a useful interpretation, such as a best-for summary or category suggestion. It is labelled as an inference, not presented as a verified claim.
• Founder-edited means the owner claimed the listing and supplied or corrected the information.

Those labels give an assistant important context. A visible price and a founder-written description are not the same kind of evidence. An inferred summary can help with discovery, but it should not silently become a product promise. This is the same honesty model we use across the honest software directory, now exposed to agents as well as people.

That distinction also helps founders. If a listing is wrong, the answer is not to bury the error under a larger block of copy. Claim the listing, correct the facts and make the source of the correction clear.

What comes next

This is a first release. We expect the useful parts to become clearer as people connect the server to real assistants and tell us where the responses are incomplete or awkward.

Next, we want to support more agent integrations, improve the shape of returned facts and make it easier to understand why a tool matched a query. We will keep the same constraint while doing that: observed facts, inferences and founder edits should remain distinguishable.

We also want to learn whether agents need more than search and comparison. A directory might eventually expose availability, alternatives, category summaries or other views, but we would rather add a small number of useful tools than publish a large surface that is hard to trust.

Questions founders are asking

Does an MCP server guarantee that AI assistants recommend my tool?

No. It creates a public way for compatible assistants to query directree. Recommendations still depend on the user's question, the assistant, its retrieval choices and the quality and relevance of the available listings.

Do I need an MCP integration to get listed?

No. You can submit a tool normally. MCP is an access layer for assistants, not a new submission requirement.

What should I do if my listing has outdated facts?

Review the public listing, claim it if you own the product and correct the relevant fields. Accurate, clearly sourced information is more useful to people and agents than longer promotional copy.

The simplest next step is to list your software on directree. Submission is free, the listing is structured, and a claimed listing can be kept current as your product changes.