generated: '2026-08-14' method: derived source: >- Derived by aligning the 11 published MCP tools (https://help.tracxn.com/en/articles/15131870-available-data-tools-with-tracxn-mcp, captured in mcp/tracxn-mcp.yml) with the REST surface published in Tracxn's official public Postman workspace (https://www.postman.com/tracxnapi/tracxn-api), saved verbatim at postman/tracxn-api-production.postman.json — 42 requests over 35 distinct endpoint URLs across API v2.2 and v3.0. purpose: >- Make the MCP tool the first-class discovery unit and bind it to the REST endpoint that backs it. Tracxn is a two-surface provider — REST and MCP — projecting one private-market data core. The surfaces are close but NOT identical: MCP adds an entity-resolution tool and a prompt-loading tool that have no REST equivalent, while REST exposes taxonomy, aggregation and saved-list endpoints that no MCP tool reaches. surfaces: rest: contract: postman/tracxn-api-production.postman.json kind: postman-collection openapi: null openapi_note: >- Tracxn publishes NO OpenAPI. The binding target below is therefore an endpoint PATH, not an operationId — there are no operationIds to bind to. No OpenAPI was synthesised from the collection: doing so would credit Tracxn with a spec it does not publish. base_production: https://platform.tracxn.com/api/3.0 base_legacy: https://platform.tracxn.com/api/2.2 base_playground: https://platform.tracxn.com/api/2.2/playground auth: accessToken request header (see authentication/tracxn-authentication.yml) gated: false gated_note: >- The collection is public and readable anonymously; only CALLING the endpoints requires a token. Full reference docs (Output Keys, Sorts applicable) sit behind the platform login. mcp: url: https://platform.tracxn.com/mcp auth: oauth gated: true gated_note: >- tools/list requires an authenticated session, so tool inputSchemas are not readable. Tool names and descriptions below come from Tracxn's published documentation. graphql: null # Each MCP tool -> the REST endpoint(s) that back it. Because the MCP input schemas are # gated, bindings are matched on name + published description + entity semantics; the # confidence column states that honestly rather than implying a verified wire-level mapping. crosswalk: - tool: search_companies category: search rest: - POST /api/3.0/companies - POST /api/2.2/companies - POST /api/2.2/companies/search binding: rest confidence: high note: >- "Search and fetch detailed company information by name or domain" maps onto the Companies endpoint, whose filter body is the same name/domain criteria. v2.2 additionally splits a dedicated /companies/search name lookup. - tool: search_funding_rounds category: search rest: - POST /api/3.0/fundingrounds - POST /api/2.2/transactions binding: rest confidence: high note: The v2.2 entity was "transactions"; v3.0 renamed it "fundingrounds". - tool: search_acquisitions category: search rest: - POST /api/3.0/acquisitions - POST /api/2.2/acquisitiontransactions binding: rest confidence: high - tool: search_investors category: search rest: - POST /api/2.2/investors binding: rest confidence: high note: >- No v3.0 investor SEARCH endpoint appears in the production collection — only /api/3.0/investors/metric (aggregation). Investor search is still v2.2-only on REST while the MCP tool exposes it regardless of version. Worth raising with Tracxn given the announced v2.2 deprecation. - tool: search_ipo_events category: search rest: - POST /api/3.0/ipos binding: rest confidence: high note: MCP-reachable and v3.0-only; there is no v2.2 IPO endpoint. - tool: search_time_series category: timeseries rest: - POST /api/3.0/timeseries/revenue - POST /api/3.0/timeseries/netprofit - POST /api/3.0/timeseries/ebitda - POST /api/3.0/timeseries/valuation - POST /api/3.0/timeseries/balancesheetassets - POST /api/3.0/timeseries/balancesheetequity - POST /api/3.0/timeseries/balancesheetliability - POST /api/3.0/timeseries/cashflowoperatingactivity - POST /api/3.0/timeseries/cashflowinvestingactivity - POST /api/3.0/timeseries/employeecountlabourfilings - POST /api/3.0/timeseries/employeecountannualreport binding: rest confidence: medium note: >- One MCP tool ("historical data — financials and employee growth") fans out to eleven REST endpoints, one per metric. Which metric a call resolves to is server-side, so the mapping is one-to-many and cannot be pinned per-invocation without the input schema. - tool: search_sectors category: taxonomy rest: - POST /api/3.0/feeds - POST /api/3.0/practiceareas - POST /api/2.2/feeds - POST /api/2.2/practiceareas binding: rest confidence: medium note: >- Tracxn's sector taxonomy is two-level — Practice Area above Feed — and the single MCP tool spans both. Note Tracxn states the taxonomy ITSELF is proprietary and is not exposed as data; these endpoints filter by it rather than export it. - tool: search_legal_entities category: search rest: [] binding: none confidence: low note: >- Legal entities are a documented platform database with a full help-centre section, and the MCP exposes a dedicated tool for them, but NO legal-entity endpoint appears in the public production collection. Either the REST endpoint is undocumented publicly or the tool composes company endpoints server-side. Recorded as unbound rather than guessed. - tool: search_locations category: filter rest: [] binding: parameter confidence: medium note: >- No /locations endpoint exists. On REST, geography is a FILTER field (e.g. "country": ["India"] in the published request bodies), not a resource. The MCP promotes a filter to a first-class tool — a genuine surface divergence, not a missing endpoint. mcp_only: - tool: load_prompts reason: >- Loads Tracxn's own instructions and prompt templates into the client's context. This is an MCP-protocol affordance (prompt distribution) with no REST analogue and no data payload; it is also the one tool that shapes how every other tool gets used. - tool: resolve_entities reason: >- Resolves a company website/domain to Tracxn's internal entity ID. No public REST endpoint performs domain->ID resolution as a distinct operation; on REST a caller filters /companies by domain and reads the id off the result. Exposing resolution as a dedicated tool is the single most agent-relevant addition in the MCP surface, because every other tool is keyed on that internal ID. rest_only: - rest: POST /api/3.0/companies/metric reason: Aggregation/roll-up (counts and buckets over a filter) — no MCP tool returns aggregates. - rest: POST /api/3.0/fundingrounds/metric reason: Aggregation over funding rounds. - rest: POST /api/3.0/investors/metric reason: Aggregation over investors. - rest: POST /api/3.0/acquisitions/metric reason: Aggregation over acquisitions. - rest: POST /api/2.2/bookmarkslist reason: >- Custom/bookmark list retrieval — a per-user platform workspace object rather than market data. No MCP tool reaches saved lists, sourcing boards or saved searches. - rest: POST /api/2.2/companies/timeseries/ebidta reason: >- Legacy v2.2 time-series variant (note the provider's "ebidta" spelling, corrected to "ebitda" in v3.0). Superseded rather than genuinely REST-only. coverage: mcp_tools: 11 mcp_tools_bound_to_rest: 7 mcp_tools_unbound: 4 # load_prompts, resolve_entities, search_legal_entities, search_locations rest_endpoints_distinct: 35 rest_endpoints_reachable_via_mcp: 22 rest_endpoints_mcp_unreachable: 6 input_schemas_available: false divergence_summary: >- Neither surface is a superset. MCP wins on entity resolution and prompt guidance and flattens the v2.2/v3.0 version split away from the caller entirely; REST wins on aggregation (the four /metric endpoints), on saved lists, and on addressing a specific financial metric directly instead of through one catch-all time-series tool. An agent restricted to MCP cannot compute a market count without paying credits to enumerate results, which is the most consequential gap in the crosswalk.