ChimeraLang MCP Server — probabilistic types, consensus gates, and hallucination detection as Claude tools
chimeralang-mcp
Give Claude typed confidence, hallucination detection, and constraint enforcement — as native MCP tools.
ChimeraLang is a programming language built for AI cognition. This MCP server exposes its runtime as 9 tools Claude can call during any conversation — no Anthropic permission needed, works today with Claude Desktop and Claude Code.
Install
pip install chimeralang-mcp
# or
uvx chimeralang-mcp
Claude Desktop Setup
Add to your config file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"chimeralang": {
"command": "uvx",
"args": ["chimeralang-mcp"]
}
}
}
Or with pip-installed version:
{
"mcpServers": {
"chimeralang": {
"command": "python",
"args": ["-m", "chimeralang_mcp"]
}
}
}
Restart Claude Desktop — 9 ChimeraLang tools are now available.
Tools
| Tool | What it does |
|---|---|
| chimera_run | Execute a .chimera program string |
| chimera_confident | Assert a value meets >= 0.95 confidence threshold |
| chimera_explore | Wrap a value as exploratory (hallucination explicitly permitted) |
| chimera_gate | Collapse multiple candidates via consensus (majority / weighted_vote / highest_confidence) |
| chimera_detect | Hallucination detection — 5 strategies: range, dictionary, semantic, cross_reference, temporal |
| chimera_constrain | Full constraint middleware on any tool result |
| chimera_typecheck | Static type-check a .chimera program |
| chimera_prove | Execute + Merkle-chain integrity proof |
| chimera_audit | Session-level call log and confidence summary |
What problem does this solve?
Claude's tool-use loop has no built-in mechanism for:
- Confidence gating — only proceed if confidence >= threshold
- Typed output contracts — this result must satisfy constraint X before going downstream
- Genuine consensus detection — is multi-path agreement real, or trivially identical?
- Hallucination signals — structured detection, not just "does it sound right"
- Trust propagation — confidence degrades through chained tool calls; nothing tracks it
ChimeraLang fills exactly these gaps as a constraint layer sitting between Claude and its tools.
Example prompts
Gate a value before a critical action:
"Before you submit that form, use chimera_confident to verify you're >= 0.95 confident the data is correct."
Consensus across reasoning paths:
"Generate 3 different answers, then use chimera_gate with weighted_vote to collapse to the most reliable one."
Hallucination scan on output:
"After you get that search result, run chimera_detect with semantic strategy to check for absolute-certainty markers."
Full constraint pipeline:
"Use chimera_constrain on that tool result with min_confidence 0.85 and detect_strategy semantic."
Integrity proof for audit:
"Run this reasoning with chimera_prove so we have a tamper-evident trace."
ChimeraLang Quick Reference
// Confident<> — enforces >= 0.95 confidence
val answer: Confident<Text> = confident("Paris", 0.97)
// Explore<> — hallucination explicitly permitted
val hypothesis: Explore<Text> = explore("maybe dark matter is...", 0.4)
// Gate — multi-branch consensus
gate verify(claim: Text) -> Converge<Text>
branches: 3
collapse: weighted_vote
threshold: 0.80
return claim
end
// Detect — hallucination scan
detect temperature_check
strategy: "range"
on: temperature
valid_range: [-50.0, 60.0]
action: "flag"
end
Links
- ChimeraLang core: github.com/fernandogarzaaa/ChimeraLang
- OpenChimera: github.com/fernandogarzaaa/OpenChimera_v1
License
MIT © Fernando Garza