diff --git a/docs/README.md b/docs/README.md index fc86d4c..4359c1a 100644 --- a/docs/README.md +++ b/docs/README.md @@ -106,5 +106,3 @@ Use an **anti-hype, high-density, clinical tone**: - **Spec and product discussion:** [agentfinder-project/agentfinder issues](https://github.com/agentfinder-project/agentfinder/issues) - **Documentation changes:** Open a PR in this repository (`agentfinder-project/docs`). Public write access may be limited; ask in issues or [Discord](https://discord.gg/PK3DpEybzP) if you need collaborator access. - -See also [Contributors](contributors.md) on the published site. diff --git a/docs/anthropic_mcp.md b/docs/anthropic_mcp.md deleted file mode 100644 index 54e5a63..0000000 --- a/docs/anthropic_mcp.md +++ /dev/null @@ -1,33 +0,0 @@ -# Anthropic's MCP Registry - -!!! danger "TODO: Review Ingestion Theory" - Guha/Team: Please review the ingestion and superset mapping theory for this ecosystem integration before finalized publication. - -Agent Finder acts as an open, protocol-agnostic discovery envelope designed to ingest, wrap, and index the **Model Context Protocol (MCP)** tool ecosystem. - ---- - -## The Envelope Mapping - -The Model Context Protocol (MCP) standardizes model-to-data/tool connections. Agent Finder wraps MCP tool server endpoints under the standard `application/mcp-server+json` media type: - -```json -{ - "identifier": "urn:ai:acme.com:server:weather", - "displayName": "Acme Weather Telemetry Server", - "type": "application/mcp-server+json", - "url": "https://api.acme.com/mcp/weather.json", - "capabilities": ["WeatherTool", "ForecastTool"], - "representativeQueries": [ - "what is the current wind speed in Chicago", - "get the 5-day forecast for Seattle" - ] -} -``` - ---- - -## Developer Value - -* **Federated Scaling**: Developers of private or enterprise MCP tools can bypass the centralized `modelcontextprotocol.io` registry altogether and host their servers sovereignly on their own FQDNs. -* **Zero Prompt Bloat**: By indexing `representativeQueries` outside the prompt, clients (like Claude Code) can perform semantic vector searches on the registry first, feeding Claude only the top matching tool schemas dynamically at runtime. diff --git a/docs/comparisons.md b/docs/comparisons.md index 90fdd14..5b991a3 100644 --- a/docs/comparisons.md +++ b/docs/comparisons.md @@ -12,24 +12,8 @@ Agent Finder is designed as a **superset** of existing agent and tool discovery ## What about... -### What about MCP Registry - -The entire list of public MCP servers can be indexed natively. The official MCP Registry (`registry.modelcontextprotocol.io`) can remain exactly as it is today, while also acting as a federated source of truth that is crawled and made queryable via Agent Finder search endpoints. - -`TODO(Guha): please fill in` - ---- - ### What about ACP Agent Registry The list of ACP Agents in [ACP's Agent Registry](https://agentclientprotocol.com/get-started/registry) is already structurally close to the AI Catalog specification. ACP registries can easily export their directory manifests as standard `ai-catalog.json` feeds, enabling instant web-scale discovery for editor-context tools. -`TODO(Guha): please fill in` - ---- - -### What about Open Plugins - -A plugin under the [Open Plugins](https://open-plugins.com/) standard is essentially a bundle of MCP servers, Skills, or other tools. By describing these bundles inside the standard AI Catalog specification, Open Plugins can leverage Agent Finder's federated infrastructure for instant decentralized deployment, crawling, and semantic indexing. - -`TODO(Guha): please fill in` +`TODO(Guha): please fill in` \ No newline at end of file diff --git a/docs/contributors.md b/docs/contributors.md deleted file mode 100644 index 465bcd8..0000000 --- a/docs/contributors.md +++ /dev/null @@ -1,27 +0,0 @@ -# Contributors - -## Core Team - -This was designed and reviewed by a core team of agent builders at: - -- Amazon -- Cisco -- GoDaddy -- Google -- HuggingFace -- Microsoft - -All of you are building agents, tools, skills, and more. We encourage you to add support for AI Catalog and Agent Finder—see our [How to publish](how_to_publish.md) guide. - -## Contributors & Endorsers - -- Anthropic (pending confirmation) -- Nvidia (pending confirmation) -- SAP (pending confirmation) -- Snowflake (pending confirmation) - -## How to contribute - -Right now PRs are not publicly available. - -Please communicate in [issues](https://github.com/agentfinder-project/agentfinder/issues) or our [Discord](https://discord.gg/PK3DpEybzP) to discuss your ideas, and we can grant PR access. diff --git a/docs/how_to_build_search.md b/docs/how_to_build_search.md index 876902e..fd04fa8 100644 --- a/docs/how_to_build_search.md +++ b/docs/how_to_build_search.md @@ -5,7 +5,7 @@ A Search Registry is an active service that crawls static manifests, indexes capabilities semantically, and exposes standard REST search endpoints (`POST /search`) to clients. -Building your own Search Registry is entirely optional. If you only want to publish capabilities, you only need to host a static manifest (see [How to Publish](how_to_publish.md)). To query capabilities, your client can connect to any existing public or private search registry (see [Implementations](implementations.md) for active endpoints). +Building your own Search Registry is entirely optional. If you only want to publish capabilities, you only need to host a static manifest (see [How to Publish](how_to_publish.md)). To query capabilities, your client can connect to any existing public or private search registry. --- diff --git a/docs/implementations.md b/docs/implementations.md deleted file mode 100644 index a2506d3..0000000 --- a/docs/implementations.md +++ /dev/null @@ -1,52 +0,0 @@ -# Implementations - -Discover active implementations, dynamic search sandbox endpoints, and platform catalogs. - ---- - -## Reference Implementations - -
- -
- Active -

Mindpower Agent Finder

-

An example client-side implementation indexing, listing, and running search discovery for multiple registered agents and tools.

- Visit Live Directory -
- -
- Active -

HuggingFace EvalState API

-

An example dynamic search endpoint built according to the Agent Finder specification. Demonstrates direct POST /search capabilities.

- View API Docs -
- -
- ---- - -## Upcoming Enterprise Support - -We are partnering with top enterprise platforms to roll out native discovery endpoints and standard `.well-known/ai-catalog.json` support. - -
-
- Coming Soon -

Google Workspace Agent Catalog

-

Native discovery feeds and SPIFFE-attested search endpoints for Gemini Enterprise Agent Platform integrations.

-
-
- Coming Soon -

Microsoft Copilot Registry

-

Auto-discovery and tenant-isolated dynamic routing search engines matching Entra ID authorization profiles.

-
-
- ---- - -## Joining the Ecosystem - -Exposing your agents and tools to Agent Finder registries is simple. - -Please refer to our step-by-step **[How to Publish](how_to_publish.md)** guide for detailed instructions on hosting your `ai-catalog.json` manifest, setting up DNS service bindings, and enabling vector embeddings. diff --git a/docs/introduction.md b/docs/introduction.md index bf4f393..b989f2b 100644 --- a/docs/introduction.md +++ b/docs/introduction.md @@ -12,12 +12,9 @@ We have seen this before. In 1994, finding something on the web meant a bookmark ## The next step -The discovery problem is well recognized, and the work so far has come in two kinds. Toolbox (Microsoft Foundry) and MCP registries provide mechanisms to create curated sets of augments — a vetted collection a developer can point at. ToolLLM (Qin et al., 2023) and Claude's tool-search take a different step: they make selection cheaper by moving part of it out of the context window, retrieving the relevant augments per query rather than injecting all of them. +The discovery problem is well recognized, and the work so far has come in two kinds. Toolbox (Microsoft Foundry) provides a mechanism to create curated sets of augments — a vetted collection a developer can point at. ToolLLM (Qin et al., 2023) and Claude's tool-search take a different step: they make selection cheaper by moving part of it out of the context window, retrieving the relevant augments per query rather than injecting all of them. -Agent Finder takes the next logical step: **an abstraction layer over both.** It is a protocol, not a product or a particular search engine, and it pins down two things while leaving the rest open: how an augment is described, and how a client asks the discovery question and reads the answer. The service that answers can be built however its provider chooses — including with the very techniques above. This turns the existing curated sets into something a client calls uniformly, so an augment published once is found by many clients, and a client reaches augments far beyond its own pre-connected set. - -!!! note "Find, don't invoke" - Agent Finder finds, it does not invoke. Discovery returns which augment fits and where to reach it; invocation happens over that augment's own protocol. +Agent Finder takes the next logical step: **an abstraction layer over both.** It is a protocol, not a product or a particular search engine, and it pins down two things while leaving the rest open: how an augment is described, and how a client asks the discovery question and reads the answer. The service that answers can be built however its provider chooses — including with the very techniques above. This turns the existing curated sets into something a client calls uniformly, so an augment published once is found by many clients, and a client reaches augments far beyond its own pre-connected set. Agent Finder only finds; it does not invoke. Discovery returns which augment fits and where to reach it, and the client then invokes it over the augment's own protocol. --- @@ -29,8 +26,7 @@ The more consequential case is the **enterprise**. An IT or platform team runs i Because the protocol allows aggregation across endpoints, control also means **composition**. An IT manager can define their instance as "everything GitHub's Agent Finder serves, plus the augments in our internal one" — a union of a public endpoint and a private one, queried as a single view. -!!! tip "The DNS property" - This is a property DNS has and web search does not: DNS resolvers compose, forwarding to upstream servers and merging the global namespace with local zones, so every organization runs its own resolver without leaving the shared system. Web search has no equivalent — you query one engine's index, and there is no standard way to say "these results plus my company's, merged." Agent Finder inherits the DNS property: instances compose, so an enterprise extends the public ecosystem rather than walling itself off from it. +This is a property DNS has and web search does not: DNS resolvers compose, forwarding to upstream servers and merging the global namespace with local zones, so every organization runs its own resolver without leaving the shared system. Web search has no equivalent — you query one engine's index, and there is no standard way to say "these results plus my company's, merged." Agent Finder has a similar property: instances compose, so an enterprise extends the public ecosystem rather than walling itself off from it. --- @@ -40,8 +36,7 @@ Agent Finder is exposed to a client as a **Skill**, and equivalently as an **MCP The other side is **publishing**. Anyone with an augment they want discovered — a single tool or skill, or a whole internal set — describes it in a standard catalog file, hosts it on their own domain, and advertises it via a well-known URL, a `robots.txt` entry, an HTML tag, or DNS. Any Agent Finder–compatible instance can then index those augments. The catalog is published once and reachable by every instance that chooses to include it, rather than integrated separately into each client. -!!! info "Two distinct roles" - Publishing a catalog makes an augment *available* to be indexed; running an instance decides what actually gets *served* to a given set of users. +These are two distinct roles: publishing a catalog makes an augment *available* to be indexed, while running an instance decides what actually gets *served* to a given set of users. --- diff --git a/docs/open_plugins.md b/docs/open_plugins.md deleted file mode 100644 index f94bd0b..0000000 --- a/docs/open_plugins.md +++ /dev/null @@ -1,38 +0,0 @@ -# Open-Plugins Ecosystem - -!!! danger "TODO: Review Ingestion Theory" - Guha/Team: Please review the ingestion and superset mapping theory for this ecosystem integration before finalized publication. - -Agent Finder acts as an open, protocol-agnostic discovery envelope designed to ingest, wrap, and index the community **Open Plugins** standard. - ---- - -## The Envelope Mapping - -Open Plugins describe bundles of MCP servers, skills, or REST tools. Agent Finder wraps these plugin definitions natively inside the `application/ai-catalog+json` nested catalog envelope: - -```json -{ - "identifier": "urn:ai:acme.com:plugin:finance-suite", - "displayName": "Finance Tool Bundle", - "type": "application/ai-catalog+json", - "data": { - "specVersion": "1.0", - "entries": [ - { - "identifier": "urn:ai:acme.com:finance:mcp", - "displayName": "Finance Trading MCP", - "type": "application/mcp-server+json", - "url": "https://api.acme.com/mcp/finance.json" - } - ] - } -} -``` - ---- - -## Developer Value - -* **Federated Discoverability**: Open Plugins inherit Agent Finder's decentralized domain-anchored indexing without requiring manual registration in centralized directories. -* **Cryptographic Trust**: Plugin bundles inherit `trustManifest` attestation and signature verification, guaranteeing tamper-proof deployment across federated networks. diff --git a/docs/why.md b/docs/why.md index 73bd378..981dbc4 100644 --- a/docs/why.md +++ b/docs/why.md @@ -26,7 +26,7 @@ Instead of forcing the LLM to sort through the noise, Agent Finder moves selecti We are moving away from manually installed, hardcoded integrations toward dynamic runtime discovery. -| Vector | Centralized Registries (e.g., MCP Registry) | Federated Search (Agent Finder) | +| Vector | Centralized Registries | Federated Search (Agent Finder) | | :--- | :--- | :--- | | **Discovery** | Manual registration / gatekeeper approval | Dynamic crawling and indexing (SEO for agents) | | **Hosting** | Single central repository database | Self-hosted on publisher domains | diff --git a/docs/zed_acp.md b/docs/zed_acp.md deleted file mode 100644 index 5dbc136..0000000 --- a/docs/zed_acp.md +++ /dev/null @@ -1,28 +0,0 @@ -# Zed's ACP Agent Registry - -!!! danger "TODO: Review Ingestion Theory" - Guha/Team: Please review the ingestion and superset mapping theory for this ecosystem integration before finalized publication. - -Agent Finder acts as an open, protocol-agnostic discovery envelope designed to ingest, wrap, and index **Zed's Agent Client Protocol (ACP)** capabilities. - ---- - -## The Envelope Mapping - -Zed's ACP standardizes editor-to-agent UI integrations. Agent Finder wraps ACP tool endpoints under the proposed `application/acp-agent+json` media type: - -```json -{ - "identifier": "urn:ai:acme.com:editor:refactor", - "displayName": "Acme Workspace Refactoring Agent", - "type": "application/acp-agent+json", - "url": "https://api.acme.com/acp/endpoint" -} -``` - ---- - -## Developer Value - -* **Cross-Harness Discoverability**: ACP-compliant agents gain permissionless discoverability outside of the Zed editor ecosystem, making them searchable by any harness (e.g., Gemini, Claude Code). -* **Zero Prompt Bloat**: Editor agents are dynamically searched and loaded into the client's environment only when a workspace refactoring task matches the user's active intent. diff --git a/mkdocs.yml b/mkdocs.yml index 70c4430..8943441 100644 --- a/mkdocs.yml +++ b/mkdocs.yml @@ -79,7 +79,6 @@ nav: - About: - Why Agent Finder?: why.md - Comparisons: comparisons.md - - Contributors: contributors.md - FAQ: faq.md - Introduction: introduction.md - Guides: @@ -89,8 +88,3 @@ nav: - Specifications: - Agent Finder Spec: spec.md - AI Catalog Spec: ai_catalog_spec.md - - Ecosystem: - - Implementations: implementations.md - - Open-plugins: open_plugins.md - - Zed's ACP Agent Registry: zed_acp.md - - Anthropic's MCP Registry: anthropic_mcp.md