OSMP (Octid Semantic Mesh Protocol)
OSMP MCP Server
MCP server for the Octid Semantic Mesh Protocol (OSMP). Deterministic encode/decode of agentic AI instructions by table lookup. No inference at decode.
Install
pip install osmp-mcpConnect
Claude Code
claude mcp add osmp -- osmp-mcpClaude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"osmp": {
"command": "osmp-mcp"
}
}
}Cursor / VS Code / Any MCP Client
{
"servers": {
"osmp": {
"command": "osmp-mcp",
"transport": "stdio"
}
}
}Tools (8)
Tool | What it does |
| Structured fields to SAL instruction |
| SAL to structured fields (handles compound instructions) |
| DAG topology and loss tolerance analysis |
| Search the opcode dictionary by namespace and/or keyword |
| Search domain corpora by keyword and/or code prefix (ICD-10-CM, ISO 20022, MITRE ATT&CK) |
| Single domain code lookup (exact code required) |
| Multiple exact domain codes in one call |
| Canonical conformance suite |
Resources (6)
URI | Content |
| SAL grammar, composition rules, and reference (~390 tokens) |
| Protocol design philosophy |
| Full ASD (342 opcodes, 26 namespaces) |
| SAL formal grammar (EBNF) |
| Available D:PACK/BLK domain corpus stats |
| 11 annotated SAL examples |
Agent Quickstart
Read
osmp://system_promptUse
osmp_lookupto find opcodes by namespace or keywordCompose SAL directly from the dictionary
Use
osmp_discoverto find domain codes you don't know (keyword + optional prefix)Use
osmp_resolvefor exact code lookup once you have the codeUse
osmp_compound_decodeto check DAG topology before transmittingUse
osmp_decodeto parse received instructions
Context Window Footprint
~124 tokens on connect (server instructions + tool schemas + resource listings). ~514 tokens total after reading the system prompt. Composition rules added to prevent common grammar errors observed during agent testing. Still under 0.3% of a 200K context window.
Links
License
Apache 2.0 with express patent grant.
This server cannot be deployed
Maintenance
Related MCP Connectors
10 deterministic micro-tools for agents. x402 on X Layer, zero LLM.
Machine-native utility network: verified evidence services for autonomous agents.
Sovereign Agent OS — Persistent Memory, Governance & Compliance for AI Agents.
Ricardian contracts for AI agents — dual-format, SHA-256 bound, legible by construction.