tekmeria
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| NCBI_EMAIL | No | Contact address NCBI asks for; sent only when set. | |
| NCBI_API_KEY | No | Raises the PubMed limit from 3 to 10 requests per second. | |
| CROSSREF_MAILTO | No | Puts Crossref requests in its polite pool; sent only when set. | |
| OPENFDA_API_KEY | No | Raises openFDA's daily limit from 1,000 to 120,000 requests. | |
| TEKMERIA_LOG_LEVEL | No | Log level for the JSON logs on stderr (default `INFO`). | INFO |
| TEKMERIA_CACHE_PATH | No | Where the SQLite cache lives (default: your user cache folder). | |
| TEKMERIA_HANDLE_KEY | No | The handle key, in hex: at least 32 bytes (64 hex digits). By default a random key is created on first run and stored in your user data folder. | |
| TEKMERIA_TOOL_DEADLINE_S | No | Seconds each tool call may take (default 45). | 45 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
tekmeria returns biomedical evidence (articles, preprints, trial registrations and drug labels) as passages with handles such as pmid:/abstract.~. Search, then fetch_passages for citable text. Cite only handles you received, in square brackets after each claim, and check drafts with audit_answer. See the cite_by_handle prompt.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_literatureB | Search PubMed and Europe PMC (including preprints). Returns record summaries; each citable hit has its title passage's handle. |
| search_trialsA | Search ClinicalTrials.gov registrations (give a condition, intervention or terms). Returns trial summaries; fetch nct: records for citable passages. |
| search_drug_labelsB | Search openFDA drug labels; returns every matching label, newest first. Fetch setid: records for citable passages (boxed_warning, contraindications, ...). |
| fetch_passagesA | Fetch records (pmid:, pmcid:, ppr:, nct:, setid:) and their passages. Every passage carries a citable handle; articles with a DOI get Crossref's retraction status. |
| resolve_handlesB | Look up the exact passage behind each handle, and whether the handle is genuine. |
| audit_answerB | Audit an answer's citations: genuine handle, record still exists, quotes verbatim. |
| describe_contractB | The citation contract: handle grammar, verdicts, flags, metrics and instructions. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| cite_by_handle | Instructions for answering a question with handle citations. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| audit-view | Shows an audit_answer report: verdicts, flags and the cited passages. |
TDQS
Scored across 7 tools
The three search tools are cleanly separated by data source (ClinicalTrials.gov, PubMed/Europe PMC, openFDA), and describe_contract/audit_answer occupy distinct meta and verification roles. The only mild overlap is between fetch_passages (bulk fetch records and passages by id) and resolve_handles (single-passage lookup plus genuineness check), which an agent could occasionally conflate.
All seven tools follow a strict verb_noun snake_case pattern (search_trials, fetch_passages, resolve_handles, audit_answer, describe_contract). The three search_* tools form a predictable, readable family with no style deviations.
Seven tools is well within the sweet spot and each earns its place: three source searches, one fetch, two citation-verification tools, and one contract/help tool. Nothing feels redundant or padding.
The surface covers the full evidence workflow: discover sources, fetch citable passages, resolve/verify handles, and audit an answer's citations, plus a contract description. Minor gaps remain (no pagination/continuation tool, no way to enumerate handles for a record beyond fetch_passages, no Crossref-only or retraction-status lookup standalone), but agents can work around these.