AgentFinger
| AgentFinger | |
|---|---|
| Developer | secondtruthLabs |
| Status | Draft |
| First published | 2025 |
| Latest version | 0.3 |
| Based on | CTRP |
| Transport | TCP |
| Default port(s) | 7922 (TCP), 7923 (TLS) |
| Encoding | UTF-8 |
| License | Public Domain |
AgentFinger is an application protocol for discovering and retrieving structured information about users, AI agents, organizations, and MCP servers. Built on CTRP, it is specifically optimized for consumption by Large Language Models.
Overview[edit | edit source]
AgentFinger extends the classic Finger protocol concept for the AI agent ecosystem. It provides a standardized way for agents to discover information about other agents, their capabilities, and their organizational context.
Design philosophy[edit | edit source]
- Agent-first design – Optimized for LLM consumption
- Markdown-native – Semantic readability without complex parsing
- Distributed architecture – No central registry required
- Reasonable security – Designed for trusted networks
Technical specification[edit | edit source]
Entity types[edit | edit source]
AgentFinger supports four entity types:
| Type | Prefix | Description |
|---|---|---|
| User | user: |
Human users and their profiles |
| Organization | org: |
Organizations and projects |
| Agent | agent: |
AI agents and their capabilities |
| MCP Server | mcp: |
MCP server capabilities |
Query format[edit | edit source]
Queries follow a URI-like structure:
{{{code}}}
Examples:
{{{code}}}
Type prefixes are mandatory to avoid ambiguity.
Response format[edit | edit source]
Responses use Markdown with YAML frontmatter:
{{{code}}}
Path-based resources[edit | edit source]
Entities can expose hierarchical resources:
{{{code}}}
Query parameters[edit | edit source]
| Parameter | Values | Description |
|---|---|---|
format |
md, json, yaml |
Response format |
section |
Section name | Return only specified section |
Security considerations[edit | edit source]
AgentFinger is designed for trusted network environments. Implementations should consider:
- Access control lists for sensitive information
- Rate limiting to prevent abuse
- TLS encryption for sensitive data
- Authentication for write operations (if supported)
Discovery[edit | edit source]
AgentFinger services can be discovered using TSDP:
{{{code}}}
Implementations[edit | edit source]
Reference implementation[edit | edit source]
The reference server implementation is written in Go.
Use cases[edit | edit source]
- Agent-to-agent communication and discovery
- Identity and context discovery for LLM personalization
- Organization and project documentation
- MCP server capability advertisement
See also[edit | edit source]
- Common Text-based Response Protocol
- Trivial Service Discovery Protocol
- wp:Model_Context_Protocol
- wp:Finger protocol
References[edit | edit source]