spawn - The MCP Server That Builds MCP Servers

What is spawn?

Every AI agent session produces patterns: recurring tool sequences, multi-step workflows, domain-specific conventions. These patterns are valuable but ephemeral.

spawn makes them permanent. It extracts patterns from session logs, scores them for automation potential, and generates complete MCP server packages - server.py, tests, pyproject.toml, README.

AI-agnostic. 17 tools. 156 tests. 3 dependencies.

How it works

Session logs
    -> PARSE (extract tool calls, AI actions, user requests)
    -> EXTRACT (identify recurring sequences)
    -> SCORE (rate buildability: frequency, complexity, feasibility)
    -> GENERATE (complete MCP server from Jinja2 templates)
    -> VALIDATE (verify structure and imports)
    -> LEARN (record outcomes, refine scoring)

17 MCP Tools

get_log_content

Parse session logs into structured data for AI analysis

find_recurring_themes

Frequency analysis to surface automation candidates

define_pattern

AI defines patterns identified from content

score_patterns

5-dimension buildability scoring (0-100)

generate_mcp

Generate complete MCP server from pattern

run_pipeline

Full end-to-end: analyze -> score -> generate

store_pattern

Save to persistent pattern library

learn_outcome

Record results, refine scoring weights

Plus: list_patterns, get_pattern, search_patterns, suggest_similar, generate_preview, validate_mcp, compare_existing, batch_analyze

Scoring Model

Dimension	Weight	Measures
Frequency	0.3	How often the pattern appears
Complexity	0.2	Complex enough to justify automation?
Feasibility	0.2	Implementable as an MCP server?
Impact	0.2	Value of automating this pattern
Novelty	0.1	Already served by existing tools?

Get Involved

spawn is open source under the MIT license.

View on GitHub Report Issue