Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@EVEE MCP Server search for BRCA1 variants and compare their pathogenicity predictions"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP server for the EVEE (Evo Variant Effect Explorer) API — interpretable variant effect prediction from Evo 2 genomic foundation model embeddings.
EVEE provides pre-computed pathogenicity predictions, disruption profiles, and AI-generated mechanistic interpretations for 4.2 million ClinVar variants. See Pearce et al. (2026) for the full paper.
Tool | Description |
| Autocomplete-style lookup (≤6 matches) for a gene symbol, rsID, or ClinVar variation ID; useful for finding candidate variant IDs, not an exhaustive gene-wide ranking |
| Clinical summary, EVEE-derived effect outputs, database comparison scores when present, and the AI-generated mechanistic interpretation (auto-triggers on-demand generation when not yet stored) |
| Poll EVEE's on-demand interpretation generation until it completes or times out |
| Side-by-side summary (up to 10 variants in one call) — clinical label, pathogenicity, top-1 disruption, HGVS, deep link |
| Top biological annotation disruptions ranked by magnitude, optionally scoped to a category — explains why a variant is predicted pathogenic or benign |
| Full annotation probe values (325 annotations across 13 categories), optionally filtered by category |
amino_acid, atacseq, ccre, chipseq, chromhmm, elm, fstack, protein_feature, interpro, genomic_feature, ptm, region, secondary_structure
Related MCP server: Evo2 MCP Server
A Claude Code skill lives at .claude/skills/evee/SKILL.md. Agents that load it get ~120 lines of guidance on how to use the tools effectively — workflow steps, reliability caveats by variant class, and gotchas (0-based coords, indel VCF anchoring, case-sensitivity).
Requires Python 3.10+ and uv.
git clone git@github.com:goodfire-ai/evee-mcp.git
cd evee-mcp
uv syncThe repo ships a .mcp.json; opening the directory triggers the permission prompt:
claudeAdd to claude_desktop_config.json:
{
"mcpServers": {
"evee": {
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/evee-mcp", "python3", "server.py"]
}
}
}Is the FBN1 variant rs1597537935 pathogenic, and why?
The agent searches, fetches the clinical summary with the AI-generated mechanistic interpretation, and pulls out the top disruption evidence.
Compare multiple variants side-by-side.
Fetches clinical label, pathogenicity score, gene, HGVS protein, consequence, and the top-1 disruption for each. Use when the user asks to contrast, rank, or compare 2+ variants, instead of looping get_variant.
Args: variant_ids: List of variant IDs in chr:pos:ref:alt format (max 10).
| Name | Required | Description | Default |
|---|---|---|---|
| variant_ids | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses what data is fetched (clinical label, pathogenicity score, etc.) and the maximum number of variants (10). No annotations provided, so description carries full burden and does so well.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise, starts with purpose, then lists fetched fields and parameter details in bullet format. Every sentence adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description is complete for a comparison tool with one parameter and no output schema. Could mention the output structure (e.g., a table or list) but not necessary.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, but the description explains the variant_ids parameter format ('chr:pos:ref:alt') and limit (max 10), adding significant meaning beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool compares variants side-by-side and lists the fetched fields. It differentiates from sibling 'get_variant' by explicitly saying to use instead of looping.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly says when to use (contrast, rank, compare 2+ variants) and when not (instead of looping get_variant), providing clear guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get comprehensive information about a specific genetic variant.
Returns clinical significance, model-derived scores from EVEE's heads (aligned to AlphaMissense, CADD, REVEL, SIFT, etc.), reference predictor scores from external databases when present, gene constraint (LOEUF), HGVS notation, disease associations, protein domains, and the AI-generated mechanistic interpretation.
If the stored interpretation isn't ready, this tool hits EVEE's on-demand
/analysis endpoint once: if generation has already completed, the fresh
interpretation is returned inline; otherwise the response carries an
interpretation = {status: queued/processing, detail: ...} entry and you
should call wait_for_variant_analysis to poll until it finishes.
Args: variant_id: Variant identifier in chr:pos:ref:alt format (e.g. "chr17:43092918:G:A" for BRCA1 ClinVar ID 41812). NOTE: EVEE uses 0-based positions; ClinVar/VCF/HGVS are 1-based. Subtract 1 from ClinVar pos for SNVs; for indels the offset varies — prefer search_variants.
| Name | Required | Description | Default |
|---|---|---|---|
| variant_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description fully discloses behavior: hitting an on-demand endpoint, returning status if analysis not ready, and suggesting polling. Also explains coordinate system differences, which is critical for correct usage.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is well-structured and front-loaded, but slightly verbose with some long sentences. However, all content is necessary given the tool's complexity, earning its space.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given no output schema, the description covers all essential aspects: return values, conditional behavior, input format, and edge cases. It is complete and preempts common questions.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, but the description provides extensive details: format (chr:pos:ref:alt), example, and a crucial note about 0-based vs 1-based positions. This adds significant meaning beyond the simple string type.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool gets comprehensive information about a genetic variant and lists specific attributes. It distinguishes from siblings by mentioning wait_for_variant_analysis and search_variants for alternative use cases.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says when to use this tool and when to use alternatives: if interpretation is queued, call wait_for_variant_analysis; for indels, prefer search_variants. Provides clear context for decision-making.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get detailed annotation probe values for a variant.
Returns the Evo 2 model's predicted annotation values for both the reference and alternate allele across 325 biological annotations. Each annotation shows ref (reference allele prediction), alt (alternate allele prediction), and delta (alt - ref).
Use this for deep analysis when you need the full picture — e.g., all chromatin marks across tissues, all amino acid probabilities, or every protein feature prediction. For a quick ranked view of what's most disrupted, use get_variant_disruptions instead.
Args: variant_id: Variant identifier in chr:pos:ref:alt format. category: Optional filter. One of: amino_acid, atacseq, ccre, chipseq, chromhmm, elm, fstack, protein_feature, interpro, genomic_feature, ptm, region, secondary_structure. Omit to get ALL annotations.
| Name | Required | Description | Default |
|---|---|---|---|
| variant_id | Yes | ||
| category | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations, but description fully explains output structure (ref, alt, delta per annotation) and parameter details. No mention of side effects or auth, but read-only nature is implicit.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Well-structured with clear paragraphs, bullet-like list for categories, and no redundant sentences. Efficiently conveys all necessary information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Covers both parameters, explains return structure (325 annotations with ref/alt/delta), provides category filter details, and references sibling tool. Complete for a read-only query tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, but description adds valuable semantics: variant_id format (chr:pos:ref:alt) and lists category options with example values, compensating fully.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states it gets 'detailed annotation probe values for a variant', specifies ref, alt, delta fields, and distinguishes from sibling get_variant_disruptions which offers quick ranked view.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says 'use this for deep analysis' and advises to use get_variant_disruptions for quick ranked view, providing clear when-to-use and alternative.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get the top biological annotation disruptions for a variant.
Shows which molecular features are most affected by the variant, ranked by magnitude of change. Each disruption shows what the Evo 2 model predicts for the reference vs. alternate allele across 325 biological annotations spanning protein structure, chromatin state, regulatory elements, splice sites, and more.
This is the key tool for understanding WHY a variant is predicted pathogenic or benign — e.g., a splice-site variant might show large disruptions in splice donor/acceptor annotations, while a missense variant might show disruptions in protein domain and secondary structure annotations.
Categories: amino_acid, atacseq, ccre, chipseq, chromhmm, elm, fstack, protein_feature, interpro, genomic_feature, ptm, region, secondary_structure.
Args: variant_id: Variant identifier in chr:pos:ref:alt format. top_n: Number of top disruptions to return (default 15, max 100). category: Optional category filter — restrict ranking to one category (e.g. to see only splice-related disruptions: category='genomic_feature').
| Name | Required | Description | Default |
|---|---|---|---|
| variant_id | Yes | ||
| top_n | No | ||
| category | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries full burden and does well: discloses it returns ranked disruptions from Evo 2 model across 325 annotations, mentions categories. However, it does not address error behavior (e.g., invalid variant_id) or provide details about output structure, but overall it is transparent about what the tool does.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Well-structured with clear sections: purpose, explanation, categories list, and parameters. Slightly verbose with the categories list and example, but all information is useful. The front-loaded sentence effectively states the action.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
While param descriptions are solid and the output concept is explained, the absence of an output schema means the description should ideally detail the returned fields or format. It only mentions 'shows which molecular features are most affected' which is vague. Could be more complete for a tool that returns structured data.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, but the description fully compensates: specifies variant_id format, top_n default and max, and category filter with an example. This adds essential meaning beyond the bare schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clearly states it retrieves top biological annotation disruptions for a variant. The explanation of ranking and categories adds clarity. However, it does not explicitly differentiate from sibling tools like get_variant_annotations, which might also provide disruption info, so purpose is clear but not fully distinguished.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides context that this is the key tool for understanding pathogenicity, and includes example usage with category filter. However, it does not specify when not to use it or compare it directly to sibling tools, leaving some ambiguity about choosing among similar tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Autocomplete-style variant lookup (up to 6 matches) in the EVEE database.
The query must be ONE of these exact types — do NOT combine them or add extra words like "pathogenic":
A gene name: "BRCA1", "TP53", "FBN1"
An rsID: "rs1597537935"
A ClinVar variation ID (numeric): "655979"
Returns at most 6 autocomplete-style matches. Pagination/limit params are ignored by the backend.
WARNING: a gene-name query returns an adjacent-position autocomplete slice of variants in that gene — NOT the top-pathogenicity variants of the gene. Do not infer the gene's pathogenic landscape from these 6 rows. To look up a specific variant, query by rsID or ClinVar variation ID.
Use this as the starting point to find variant IDs for the other tools.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description fully discloses behavioral traits: returns at most 6 matches, pagination/limit parameters are ignored, and the nature of gene-name query results (adjacent-position slice, not pathogenic landscape). It also warns about inferring pathogenic landscape from limited rows.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is well-structured with a clear first sentence stating purpose and limit, followed by bullet-like specifics for query types, and warnings. Every sentence adds value without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool has an output schema (not shown but indicated), the description does not need to detail return values. It fully covers behavior, limitations, and usage context. For a single-parameter search tool, it is highly complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema provides no description for the single 'query' parameter (0% schema description coverage). The description compensates fully by explaining that the query must be one of three exact types (gene name, rsID, ClinVar variation ID) with examples, and explicitly what not to do.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool performs 'autocomplete-style variant lookup' in the EVEE database, specifies the exact query types (gene name, rsID, ClinVar ID), and notes it returns up to 6 matches. It distinguishes itself from sibling tools by indicating it is the starting point for finding variant IDs for other tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides explicit when-to-use guidance (starting point for finding variant IDs) and when-not-to (do not combine query types, do not add extra words like 'pathogenic'). It also warns that gene-name queries do not return top-pathogenicity variants and advises using rsID or ClinVar ID for specific variants.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Poll EVEE's on-demand interpretation until it completes or times out.
Use this when get_variant reports interpretation.status as queued or
processing. Returns the same curated variant summary as get_variant, plus
a wait_status entry with attempts / elapsed_seconds. If the deadline
hits before completion, call this tool again to keep polling.
Args: variant_id: Variant identifier in chr:pos:ref:alt format. timeout_seconds: Maximum wall-clock time to wait (clamped to [1, 60]). poll_interval_seconds: Delay between polls (clamped to [0.5, 10]).
| Name | Required | Description | Default |
|---|---|---|---|
| variant_id | Yes | ||
| timeout_seconds | No | ||
| poll_interval_seconds | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, but the description fully discloses polling behavior, timeout clamping, and return value details (wait_status entry). It is transparent about what happens on deadline hit.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is well-structured with clear sections for purpose, usage, return, and argument details. It is slightly verbose but efficiently uses sentences without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a polling tool with no output schema, the description explains the return behavior (same as get_variant plus wait_status), all parameters, and usage context. It is fully self-contained and complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Despite 0% schema description coverage, the description explains each parameter thoroughly: variant_id format, timeout_seconds and poll_interval_seconds with clamping ranges and defaults, adding significant meaning.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description starts with 'Poll EVEE's on-demand interpretation until it completes or times out', providing a specific verb and resource. It clearly distinguishes from sibling tools like get_variant by indicating its polling nature.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly states 'Use this when `get_variant` reports `interpretation.status` as queued or processing', including an alternative scenario. Also notes that if timeout occurs, call again, providing clear guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0compare_variantsget_variantget_variant_annotationsget_variant_disruptionssearch_variantswait_for_variant_analysisScored across 6 tools
Each tool has a clearly distinct purpose: search_variants for lookup, get_variant for comprehensive info, compare_variants for side-by-side comparison, get_variant_annotations for deep probes, get_variant_disruptions for top disruptions, and wait_for_variant_analysis for polling. No overlapping functionality.
All tool names follow the verb_noun pattern in snake_case consistently (e.g., compare_variants, get_variant_annotations). The naming is predictable and easy to understand.
With 6 tools, the server covers the core workflows of variant lookup, detailed retrieval, comparison, annotation analysis, disruption ranking, and async waiting. This is well-scoped for a specialized domain without being too few or too many.
The tool set covers essential operations: search, retrieve, compare, deep annotations, disruptions, and async polling. A minor gap is the lack of a tool to list all variants for a given gene (search_variants only returns up to 6 autocomplete results), but for targeted variant analysis, the surface is sufficiently complete.
Bioinformatics MCP for genomic variant interpretation, gene-disease evidence and literature.
Look up allele frequencies by ancestry, gene constraint, variants, and coverage over gnomAD.
Protein analysis: ESM-2/ESMC embeddings, mutation scoring, landscape scans, ESMFold structure.
Broad Institute gnomAD genomic variant database (GraphQL)
