Model Finder
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MODEL_FINDER_ROOT | No | The project path the server should treat as the workspace. Repository tools read only inside the workspace: the roots your client shares over MCP, else MODEL_FINDER_ROOT, else the server's working directory. If the client has no workspace variable, write the absolute project path. A filesystem root or home directory is refused. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| modelfinder_audit_modelsA | List AI model IDs (quoted literals) and AI SDKs used in the repository, with file:line locations, provider family, route hints (Bedrock, Vertex AI, Azure, OpenRouter), usage counts and whether each ID is hard-coded or an environment-variable default. Read-only; skips secret-like files and dependency folders; never returns line content. Results are sorted by usage and paginated. |
| modelfinder_draft_eval_examplesA | Draft eval examples from prompt files and fixtures (prompts/, evals/, fixtures/, *.prompt.md, and JSON/JSONL with input, prompt or chat messages). Drops any candidate with detected personal data or secrets and reports only categories and counts. Drafts have empty checks: agree checks with the user before modelfinder_create_eval_config. Paginated; ids are stable across pages. |
| modelfinder_create_eval_configA | Assemble an eval config from candidate models, examples with checks, and settings; validate it and screen the examples for personal data and secrets. Returns the config to save as JSON for eval_models.py. Keys are referenced by environment variable name only. |
| modelfinder_validate_eval_configA | Check an eval config before running: schema, no keys in the config, HTTPS endpoints, no template placeholders, valid checks. Reports API call count, required environment variable names, unpriced models, and whether the personal-data guard would block the run. Makes no network calls. |
| modelfinder_check_outputA | Run deterministic checks on one model reply. Use it to design eval checks and routing guards before spending money on a run. |
| modelfinder_rank_eval_resultsA | Read a results.json object and return the cheapest model that met the pass threshold, ranked by cost per passing call, plus the cost/reliability cascade and models dropped as dominated. |
| modelfinder_generate_routingA | Turn a results.json object into a cheapest-first routing.json (with guards and a personal-data policy) and a LiteLLM proxy config. Returns both; nothing is written or deployed. |
| modelfinder_estimate_costA | Planning estimate of monthly token cost from requests, average tokens per request and USD-per-million prices, with optional cached-input share and pass rate (cost per accepted result, counting retries). Prices must come from the provider's official pricing page. |
| modelfinder_estimate_memoryA | Estimate memory to run an open-weight model locally: weights at a precision plus KV cache for a context length and concurrency (standard MHA/GQA attention only), with headroom against available memory. Give all four of layers, kv_heads, head_dim and context to include the KV cache. |
| modelfinder_scan_sensitive_dataA | Detect Aadhaar (Verhoeff checked), PAN, payment cards (Luhn checked), email addresses, UPI IDs, Indian mobile numbers and API or private keys in text. Returns categories and counts only, never the values. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 10 tools
Each tool targets a distinct stage or artifact (audit, draft examples, create/validate config, check output, rank results, generate routing, cost/memory estimates, sensitive-data scan). The only mild overlap is between cost-related tools (rank_eval_results vs estimate_cost) and config-checking (validate_eval_config vs check_output), but the descriptions clearly separate config validation from single-reply checking.
All ten tools share the modelfinder_ prefix and use a consistent snake_case verb_noun pattern (audit_models, draft_eval_examples, create_eval_config, rank_eval_results, etc.). No convention mixing or vague names.
Ten tools is well within the ideal range and each maps to a concrete step in the model-eval-and-routing workflow, so none feel redundant or padded.
The surface covers the full planning pipeline: auditing model usage, drafting examples, building/validating configs, checking outputs, ranking results, generating routing, and estimating cost/memory, plus PII scanning. The only gap is that actual eval execution is delegated externally (config is saved for eval_models.py), leaving a slight end-to-end dead end.